Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

The location of the Italian manufacturing industry, 1871–1911: a sectoral analysis

The location of the Italian manufacturing industry, 1871–1911: a sectoral analysis Abstract This study focuses on industrial location in Italy during the period 1871–1911, when manufacturing moved from artisanal to factory-based production processes. There is general agreement in the historical and economic literature that factor endowment and domestic market potential represented the main drivers of industrial location. We test the relative importance of the above drivers of location for the various manufacturing sectors using data at the provincial level. Estimation results reveal that the location of capital intensive sectors (such as chemicals, cotton, metalmaking and paper) was driven by domestic market potential and literacy. Once market potential and literacy are accounted for, the evidence on the effect of water endowment on industrial location is mixed, depending on the manufacturing sector considered. 1. Introduction Economic theory suggests that factor endowment and market access are the key determinants of industrial location. On the one hand, the neoclassical trade theory predicts that regional differences in factor endowments (such as mineral deposits, water supply and labor skills) contribute to determine regional comparative advantages and, therefore, regional specialization. On the other hand, the New Economic Geography (NEG) literature (Krugman, 1991; Fujita et al., 1999) suggests that an uneven spatial distribution of market access encourages firms to concentrate in regions with higher market potential, to benefit from increasing returns and to export goods and services to other regions. The relative effect of factor endowment and market access on industrial location can be hardly quantified using empirical data referring to modern economies, as the two forces tend to coexist and interact in a very complex way along with the effect of (endogenous) policy interventions. Midelfart-Knarvik and Overman (2002) show for instance that the European industrial policy strongly influenced the industrial location patterns across the EU regions. As long as industrial policies are endogenously driven by the actual spatial distribution of economic activities, it can be quite difficult to quantify the genuine (net of industrial policies) effect of comparative advantages and/or market potential on industrial location. However, the use of historical data covering the years following the political unification of a country (i.e., when the process of domestic market integration was taking its first steps, and no systematic regional industrial policy existed) may provide an opportunity to better contrast the different explanations for the spatial concentration of industry and, in particular, to appreciate the role of the home-market effect. Industrialization processes, the fall in transport costs and the integration of domestic markets may indeed generate the agglomeration forces that change the distribution of economic activities across space and reinforce spatial disparities over time. As a matter of fact, the economic history literature has increasingly drawn on NEG models to analyze national historical experiences and, due to its rising importance, the manufacturing sector has received most of the attention. Examples of studies in this direction are Wolf (2007) for Poland, Klein and Crafts (2012) for the USA, Rosés (2003), Tirado et al. (2002), and Martinez-Galarraga (2012) for Spain and Crafts and Mulatu (2005) for Britain. As far as Italy is concerned, A’Hearn and Venables (2013) explore the interactions between external trade and regional disparities since the unification of the country (in 1861). The authors argue that the economic superiority of Northern regions over the rest of the country was initially based on natural advantages (in particular the endowment of water), while from the late 1880s onwards domestic market access became a key determinant of industrial location, inducing dynamic industrial sectors to locate in regions with a large domestic market—that is, in the North. From 1945 onwards, however, with the gradual process of European integration, foreign market access became the decisive factor; and the North, once again, had the advantage of proximity to these markets. While we broadly agree with the periodization proposed by A’Hearn and Venables (2013), we believe that it requires further qualification. Specifically, in line with Rosés (2003) for Spain and Klein and Crafts (2012) for the USA, we believe that the relative importance of factor endowment and market potential for industrial location varies according to the technology prevailing in the various sectors. In particular, market potential is expected to be more important in industries characterized by increasing returns to scale (typically, high and medium capital intensive industries), while factor endowment should be more relevant in the remaining manufacturing sectors. On the basis of these considerations, we analyze the spatial location patterns of the various branches of the Italian manufacturing industry during the period 1871–1911. Specifically, we assess the relative importance of factor endowment (water abundance and labor skills) and domestic market potential for industrial location behavior in the early phases of Italian industrialization, distinguishing among the various manufacturing sectors. Our analysis is based on sectoral value added data at 1911 prices at the provincial level, recently produced by Ciccarelli and Fenoaltea (2013, 2014). Our results clearly show that as transportation costs decreased and institutional barriers to domestic trade were eliminated, Italian provinces became more and more specialized and manufacturing activity became increasingly concentrated in a few provinces, mostly belonging to the North-West. The estimation results corroborate the hypothesis that both comparative advantages and domestic market potential have been responsible for this process of spatial concentration. However, the econometric analysis also reveals a heterogeneous reaction of the location of various branches of manufacturing to domestic market potential and factor endowment. In particular, the location of medium and high capital intensive sectors (including chemicals, cotton, metalmaking, machinery and paper) was mainly driven by the domestic market potential and literacy. Water endowment represented an important driver of industrial location for sectors of heterogeneous nature (such as chemicals, silk and leather). The rest of the paper is organized as follows. Section 2 illustrates the spatial distribution of manufacturing industry over the sample period. Section 3 describes the hypotheses on the key drivers of industrial location. Section 4 reports the estimation results. Section 5 concludes. 2. The spatial diffusion of manufacturing activity in Italy: 1871–1911 2.1. Setting the scene Figure 1 illustrates the division of the Italian territory that roughly prevailed during the 1815–1860 period. The seven pre-unitarian states (Kingdom of Sardinia, Kingdom of Lombardy-Venetia, Duchy of Parma and Piacenza, Duchy of Modena and Reggio, Grand Duchy of Tuscany, Papal States and Kingdom of the Two Sicilies) were characterized by extremely different institutions and economic policies. The coin, monetary regimes and trade policies were different. Primary schooling was mandatory only in certain pre-unitarian states, mainly those in the North. Figure 1 View largeDownload slide Italian provinces at 1911 borders grouped into pre-unitarian states (1815–1860 ca.). Figure 1 View largeDownload slide Italian provinces at 1911 borders grouped into pre-unitarian states (1815–1860 ca.). Italy was unified in 1861, although Venetia and Latium were annexed to the country only in 1866 and 1870, respectively. Between 1861 and 1870, the national capital was moved from Turin to Florence and, finally, to Rome. Soon after the political unification, policy makers realized that there was an urgent need of statistical information. Decennial population censuses were established, and dozens of annual reports to the Italian Parliament and other official publications concerning the main economic sectors (public budget and taxation, international trade, railroads, public school system) were regularly produced. The new official historical statistics divided the Italian territory into 16 regions (compartimenti, roughly NUTS-2 units) and 69 provinces (province, roughly NUTS-3 units). The borders of these administrative units (shown in Figure 1 and in the Appendix) did not change between 1871 and 1911. 2.2. The structure of economic activity Between 1871 and 1911, the composition of Italy’s GDP changed considerably, with industry increasing (from about 15% to 25%) at the expense of agriculture. At the same time, a considerable sectoral reallocation within manufacturing occurred. Table 1 reports the sectoral distribution of manufacturing value added at benchmark years (1871, 1881, 1901, and 1911), when population censuses were taken. With few exceptions, sectors tied to the production of consumption goods (roughly those from 2.1 foodstuffs to 2.6 leather) show a constant reduction in their shares. Sectors tied to the production of durable goods (roughly those from 2.7 metalmaking to 2.11 paper) follow an opposite long-term trend, with a rapid acceleration in the 1901–1911 decade. The last column of Table 1 summarizes the 1871–1911 trends, with numbers below 1 indicating a reduction of the sectoral share, and vice versa. Table 1 Manufacturing sectors: value added sharesa (percentages) Sectors 1871 1881 1901 1911 1911/1871 2.1 Foodstuffs 33.5 30.4 25.4 21.5 0.6 2.2 Tobacco 1.6 1.3 0.9 0.7 0.4 2.3 Textiles 10.3 10.3 12.8 11.1 1.1     2.3.1 Cotton 1.3 1.9 5.2 4.8 3.7     2.3.2 Wool 1.8 2.0 2.4 2.3 1.3     2.3.3 Silk 3.9 3.8 4.3 3.3 0.8     2.3.4 Other natural fibers 3.2 2.5 0.9 0.7 0.2 2.4 Clothing 6.9 7.4 6.8 6.3 0.9 2.5 Wood 10.0 9.3 9.7 10.0 1.0 2.6 Leather 10.5 11.5 11.4 7.8 0.7 2.7 Metalmaking 0.6 1.0 1.7 3.1 5.2 2.8 Engineering 17.5 17.8 18.7 21.5 1.2     2.8.1 Shipbuilding 2.0 1.4 3.0 1.9 1.0     2.8.2 Machinery 2.9 4.3 7.4 10.5 3.6     2.8.3 Blacksmith 12.5 12.2 9.1 9.0 0.7 2.9 Non-metallic mineral products 3.6 4.2 4.2 6.6 1.8 2.10 Chemicals and rubber 2.1 2.4 3.0 4.3 2.0 2.11 Paper 2.7 3.5 4.8 6.3 2.3 2.12 Sundry 0.7 0.7 0.6 0.7 1.0 2. Total manufacturing 100.0 100.0 100.0 100.0 1.0 Sectors 1871 1881 1901 1911 1911/1871 2.1 Foodstuffs 33.5 30.4 25.4 21.5 0.6 2.2 Tobacco 1.6 1.3 0.9 0.7 0.4 2.3 Textiles 10.3 10.3 12.8 11.1 1.1     2.3.1 Cotton 1.3 1.9 5.2 4.8 3.7     2.3.2 Wool 1.8 2.0 2.4 2.3 1.3     2.3.3 Silk 3.9 3.8 4.3 3.3 0.8     2.3.4 Other natural fibers 3.2 2.5 0.9 0.7 0.2 2.4 Clothing 6.9 7.4 6.8 6.3 0.9 2.5 Wood 10.0 9.3 9.7 10.0 1.0 2.6 Leather 10.5 11.5 11.4 7.8 0.7 2.7 Metalmaking 0.6 1.0 1.7 3.1 5.2 2.8 Engineering 17.5 17.8 18.7 21.5 1.2     2.8.1 Shipbuilding 2.0 1.4 3.0 1.9 1.0     2.8.2 Machinery 2.9 4.3 7.4 10.5 3.6     2.8.3 Blacksmith 12.5 12.2 9.1 9.0 0.7 2.9 Non-metallic mineral products 3.6 4.2 4.2 6.6 1.8 2.10 Chemicals and rubber 2.1 2.4 3.0 4.3 2.0 2.11 Paper 2.7 3.5 4.8 6.3 2.3 2.12 Sundry 0.7 0.7 0.6 0.7 1.0 2. Total manufacturing 100.0 100.0 100.0 100.0 1.0 aThe table includes 12 manufacturing sectors (numbered from 2.1 to 2.12, as it is customary in the national account, where usually 1 refers to the extractive sector, 2 to manufacturing, 3 to constructions and 4 to the utilities). Textiles (2.3) and engineering (2.8) are further disaggregated, as detailed in the Appendix. Numbers need not to add due to rounding. Table 1 Manufacturing sectors: value added sharesa (percentages) Sectors 1871 1881 1901 1911 1911/1871 2.1 Foodstuffs 33.5 30.4 25.4 21.5 0.6 2.2 Tobacco 1.6 1.3 0.9 0.7 0.4 2.3 Textiles 10.3 10.3 12.8 11.1 1.1     2.3.1 Cotton 1.3 1.9 5.2 4.8 3.7     2.3.2 Wool 1.8 2.0 2.4 2.3 1.3     2.3.3 Silk 3.9 3.8 4.3 3.3 0.8     2.3.4 Other natural fibers 3.2 2.5 0.9 0.7 0.2 2.4 Clothing 6.9 7.4 6.8 6.3 0.9 2.5 Wood 10.0 9.3 9.7 10.0 1.0 2.6 Leather 10.5 11.5 11.4 7.8 0.7 2.7 Metalmaking 0.6 1.0 1.7 3.1 5.2 2.8 Engineering 17.5 17.8 18.7 21.5 1.2     2.8.1 Shipbuilding 2.0 1.4 3.0 1.9 1.0     2.8.2 Machinery 2.9 4.3 7.4 10.5 3.6     2.8.3 Blacksmith 12.5 12.2 9.1 9.0 0.7 2.9 Non-metallic mineral products 3.6 4.2 4.2 6.6 1.8 2.10 Chemicals and rubber 2.1 2.4 3.0 4.3 2.0 2.11 Paper 2.7 3.5 4.8 6.3 2.3 2.12 Sundry 0.7 0.7 0.6 0.7 1.0 2. Total manufacturing 100.0 100.0 100.0 100.0 1.0 Sectors 1871 1881 1901 1911 1911/1871 2.1 Foodstuffs 33.5 30.4 25.4 21.5 0.6 2.2 Tobacco 1.6 1.3 0.9 0.7 0.4 2.3 Textiles 10.3 10.3 12.8 11.1 1.1     2.3.1 Cotton 1.3 1.9 5.2 4.8 3.7     2.3.2 Wool 1.8 2.0 2.4 2.3 1.3     2.3.3 Silk 3.9 3.8 4.3 3.3 0.8     2.3.4 Other natural fibers 3.2 2.5 0.9 0.7 0.2 2.4 Clothing 6.9 7.4 6.8 6.3 0.9 2.5 Wood 10.0 9.3 9.7 10.0 1.0 2.6 Leather 10.5 11.5 11.4 7.8 0.7 2.7 Metalmaking 0.6 1.0 1.7 3.1 5.2 2.8 Engineering 17.5 17.8 18.7 21.5 1.2     2.8.1 Shipbuilding 2.0 1.4 3.0 1.9 1.0     2.8.2 Machinery 2.9 4.3 7.4 10.5 3.6     2.8.3 Blacksmith 12.5 12.2 9.1 9.0 0.7 2.9 Non-metallic mineral products 3.6 4.2 4.2 6.6 1.8 2.10 Chemicals and rubber 2.1 2.4 3.0 4.3 2.0 2.11 Paper 2.7 3.5 4.8 6.3 2.3 2.12 Sundry 0.7 0.7 0.6 0.7 1.0 2. Total manufacturing 100.0 100.0 100.0 100.0 1.0 aThe table includes 12 manufacturing sectors (numbered from 2.1 to 2.12, as it is customary in the national account, where usually 1 refers to the extractive sector, 2 to manufacturing, 3 to constructions and 4 to the utilities). Textiles (2.3) and engineering (2.8) are further disaggregated, as detailed in the Appendix. Numbers need not to add due to rounding. In 1911, foodstuffs, textiles and engineering alone represented more than 50% of total value added in manufacturing. Effectively, Ciccarelli and Proietti (2013) show that these three sectors explain much of the variability of sectoral specialization at the provincial level, and thus act as a sort of ‘sufficient statistic’ for the whole manufacturing industry during 1871–1911. In addition, Fenoaltea (2016) documents how vast and heterogeneous was the engineering sectors and Rosés (2003) bases his results for Spain on the separated components of the textile industry. For these reasons, we disaggregated the data on value added for the textile and engineering sectors into sub-sectors (this was not possible for foodstuffs due to the lack of historical data). Table 1 clearly shows the importance of this disaggregation. Within the textile sector, only cotton, more suitable to mechanization than other fibers, increased substantially its value added share over time. Within engineering, the size of the machinery component increased substantially. At the same time, however, traditional blacksmith activities, despite a declining trend, accounted for about half of value added of the engineering sector even at the end of our sample period. After all, 19th-century Italy was very much an agricultural country and the maintenance of agricultural tools (such as spades, hoes and ploughs) represented a substantial part of blacksmiths’ traditional activity. 2.3. Regional specialization and geographical concentration of manufacturing activity To analyze the dynamics of regional specialization, industrial concentration and the spatial distribution of industrial activities during 1871–1911, we use disaggregated data on manufacturing value added, vik(t), at 1911 prices in province i ( i=1,…,69), sector k ( k=1,…,17)and time t ( t=1871,1881,1901,1911) and compute, for each time t the location quotient LQik=(vik/∑kvik)/(∑kvik/∑i∑kvik) (see the Appendix for details on data and indices). In 1871, the manufacturing industry mainly clustered in a few Northern provinces (Figure 2). To some extent, it was however also present in the South. Naples and Palermo, the provinces with the ancient capitals of the Kingdom of the Two Sicilies, registered in particular noticeable LQ values (respectively, 1.41 and 1.32) in line with Turin (1.35) and Genoa (1.20) that, together with Milan (1.57), formed the vertices of the so-called ‘industrial triangle’. In 1911, the North-West reinforced its dominant position, while most of the Southern provinces worsened their relative position. Figure 2 View largeDownload slide Manufacturing: choropleth maps of LQ values. Figure 2 View largeDownload slide Manufacturing: choropleth maps of LQ values. The increase in spatial inequalities in Italian manufacturing activity is also revealed by the values of the Theil index of concentration (passing from 0.03 in 1871 to 0.08 in 1911) and of the Moran’s I index of spatial autocorrelation (passing from a statically non-significant negative value of −0.04 in 1871 to a significant value of 0.05 in 1911). Moreover, the average degree of industrial specialization of Italian provinces (measured by the Krugman index) increased monotonically over the sample period (passing from 0.26 to 0.38). The Krugman index increased particularly for selected provinces of the North-West (including Turin, Milan, Genoa, Novara, Como, Cremona and Bergamo) and Tuscany (Massa Carrara, Lucca, Pisa and Leghorn) (see the Appendix). These findings are in line with the predictions of Core-Periphery NEG models (Krugman, 1991), according to which, when interregional trade costs decrease,1 industrial activities characterized by increasing returns to scale tend to concentrate in the core regions (i.e., those with higher demand; home-market effect), while the remaining (peripheral) regions suffer de-industrialization. Increased competition, in the labor market for instance, may, however, create an inverted U-shape pattern: as trade costs continue to fall, a second stage of adjustment could occur when market forces operate to undermine the core–periphery pattern and reduce regional inequalities (Krugman and Venables, 1995; Puga, 1999). However, this second stage of adjustment refers to the most recent history of globalization, so we may conclude that the evolution of the spatial distribution of the overall manufacturing activity during the period 1871–1911 is a good fit for the predictions of the core–periphery model. The spatial diffusion of the overall manufacturing activity is nevertheless the net result of heterogeneous dynamics across the 17 manufacturing branches. To analyze sectoral developments, we combine the a-spatial Theil index and the Moran’s I index in a scatterplot for 1871 and 1911, excluding tobacco and sundry (Figure 3). The vertical and horizontal dashed lines denote median values and identify four patterns in the distribution of economic activities: (a) HL (high concentration and low spatial dependence), (b) HH (high concentration and high spatial dependence), (c) LH (low concentration and high spatial dependence) and (d) LL (low concentration and low spatial dependence). Figure 3 View largeDownload slide Spatial concentration: scatterplot between the a-spatial Theil index of concentration (y-axis), and the Moran index of spatial autocorrelation (x-axis). Figure 3 View largeDownload slide Spatial concentration: scatterplot between the a-spatial Theil index of concentration (y-axis), and the Moran index of spatial autocorrelation (x-axis). Shipbuilding and leather appear the most extreme cases. Shipbuilding belongs both in 1871 and 1911 to spatial pattern a (high Theil, low Moran). This means that this industry concentrated its activity in a small number of (coastal) areas that were not close to each other. In this kind of sector, indeed, economies of scale are reached by increasing the plant size and concentrating the production in a small number of locations. Leather, both in 1871 and 1911, belongs to spatial pattern c (low Theil and high Moran). Other sectors kept their position over time. Capital intensive metalmaking belongs in 1871 and 1911 to spatial pattern b (high concentration and high spatial dependence). Traditional economic activities tied to the agricultural sector such as clothing, foodstuffs and wood, but also blacksmiths and non-metallic mineral products belong, although with some difference, to spatial pattern c (as does leather). Within textiles, silk and cotton, always characterized by above median Theil index, moved from spatial pattern a to spatial pattern b, becoming strongly agglomerated in 1911. On the contrary the other natural fibers components moved from spatial pattern b to spatial pattern a. The reallocation of the textile industry is consistent with Fenoaltea (2011) for Italy. The above evidence is also in line with the international literature. Rosés (2003) and Wolf (2007) document, respectively, for 19th-century Spain and for Poland in 1925–1937 that metalmaking and textiles were increasingly concentrated. Within the engineering sector, shipbuilding, machinery and blacksmith differ tremendously in terms of concentration, but also on the spatial autocorrelation dimension. This confirms the importance of considering further disaggregated data for this industry. One further notes that fast growing sectors (such as machinery, paper and chemicals) present in 1911 about median values of Theil and Moran. Finally, one notices a substantial continuity in the ranking of sectors along the vertical dimension (with traditional sectors like foodstuffs, clothing, blacksmiths, but also wood being the most dispersed sectors). Indeed, the ranking of the Theil indices remained very stable over time (the rank correlation between the Theil indices in 1871 and in 1911 is 0.95 with a p-value of 0.000). Interestingly, the stability in the level of concentration across industries is a pattern also found in Crafts and Mulatu (2006) for 19th-century Britain. So far we have documented the significant specialization and concentration of manufacturing industry that occurred during the sample period. In the following section, we will explore the potential drivers of this spatial evolution. 3. Drivers of industrial location The literature analyzing the main determinants of regional industrial location in the late 19th and the early 20th centuries focuses on the role played by factor endowment and market potential. Crafts and Mulatu (2006) show that patterns of industrial location in Britain were rather persistent and only marginally affected by falling transport costs during 1871–1911. Factor endowment (coal abundance) had a stronger effect on overall industrial location than proximity to markets. However, the latter was an attraction for industries with large plant size, above all shipbuilding and textiles. In addition, educated workers were an important input for the chemical but not for the textile sector. Klein and Crafts (2012) consider industrial location in the USA during 1880–1920 and show that both agglomeration mechanisms related to market potential and natural advantages influenced industrial location, although the former were increasingly important for sectors where plant size was relatively large, mainly located in the manufacturing belt. Rosés (2003) analyzes 19th-century Spain and shows that, as transport costs decreased and barriers to domestic trade were eliminated, Spanish manufacturing industry became increasingly concentrated in a few regions. Both market potential and factor endowment contribute to explain industrial location. However, the home-market effect played a key role for modern industries producing heterogeneous goods and experiencing increasing returns to scale. Wolf (2007) shows that skilled labor and inter-industry linkages have an important explanatory power for the industrial location in the case of 1925–1937 Poland. Inspired by this literature, in the remaining part of this section we provide possible measures of factor endowment and market potential and posit some hypotheses on their role as main determinants of industrial location in the case of 19th-century Italy. 3.1. Factor endowment 3.1.1. Water supply The historical literature (e.g., Cafagna, 1989) stressed the central role of natural endowment (above all, water) for industrial location in 19th-century Italy. The point was recently reiterated by Fenoaltea (2011), providing perhaps the most careful and sharp account of Italian industrialization over the 1861–1913 period. According to Fenoaltea, ‘the roots of the success of the Northern regions seem […] environmental rather than historical’ (p. 231): thus, factor endowment, more than socioeconomic variables such as social capital or institutions. Modern (or factory-based) systems of production gradually replaced artisanal production and ‘gave a strong advantage to the locations with a year-round supply of water (for power and also in the specific case of textiles, for the repeated washing of the material); and in Italy such locations abound only on the northern edge of the Po valley, where the Alpine run-off offsets the lack of summer rain’. Moreover, when analyzing the location dynamics of the textile industry during 1871–1911, Fenoaltea (2011, 232) mentions among the natural advantages of northern regions ‘the water that flowed in the rivers, the water suspended in the air […]’ and the presence of ‘mountain glaciers’. Beyond representing a source of motive power and adequate climate for textile industry (as stressed by Italy’s economic historians), rivers are in principle important determinants of industrial location of most manufacturing activities. Water may indeed work as a production input for washing, cooling, boiling and so on for both modern industries (such as chemicals) and more traditional industrial activities. As an example of the latter, consider the case of leather. Each step of the manufacturing process to obtain leather from the skins of the animals (tanning, retanning and finishing) requires a considerable amount of fresh water (Sundar et al., 2001). These arguments, and the lack of a comprehensive supporting literature on the matter, suggest that it is virtually impossible to posit specific hypotheses on the (possibly heterogeneous) response of the location of the various manufacturing sectors to water abundance. In principle, any sector should have benefited from the presence of rivers. So, we will essentially let the data speak for themselves, and assess through the econometric analysis whether water supply significantly affected industrial location net of the effect of the other variables included in the model (i.e., market potential and literacy). To measure water endowment at provincial level, the present study considers two time-invariant variables. The first one is the number of rivers in a province weighted by their economic ‘relevance’. The historical source Annuario Statistico Italiano (1886, 22–27) provides an exhaustive list of rivers flowing through the 69 Italian provinces. These rivers are ranked, with a value ranging from 1 to 5, according to the weight assigned to them by the experts of the Italian Military Geographic Institute (IGM) in a publication that culminates a research project that started in the 1960s (IGM, 2007). The importance of each river is established on the basis of ‘the length of each river and its socio-economic relevance’ (IGM, 2007, preface). In the IGM ranking, a value of 1 means ‘high relevance’, while a value of 5 means ‘low relevance’. Thus, our first measure of water endowment in each province i is the weighted average of the number of rivers in the province using (1−IGMrank/6) as weights: RIVERi=∑r∈i(1−IGMrankr/6), where r refers to each river flowing through province i. The second variable is a dummy indicating whether the province belongs to the Alpine region. Italian rivers are essentially of two types: Alpine rivers and Apennine rivers. Alpine rivers typically descend from the Alps, flowing from the North into the upper bank of the Po river. The Alps work in a sense as a sponge that absorbs water in the autumn and winter, and releases water in the spring and summer, when the glaciers melt. Alpine rivers are therefore rich in water throughout the year, while the Apennine rivers are relatively dry during the summer season—limiting the regular development of factory-based manufacturing activities that are not organized on a seasonal basis. In addition, Alpine rivers also have a higher flow rate than the Apennine rivers. As we will document in our econometric analysis, the interaction between the continuous variable RIVERi and the Alpine region dummy variable proved to be of significant help for our understanding of industrial location of certain manufacturing sectors, and supports nicely the need for industrial use of a year-round supply of water considered by Fenoaltea (2011). The provincial distribution of these water-related variables is illustrated in Figure 4. Figure 4 View largeDownload slide Water endowment. (a) Rivers and (b) Alpine region. Panel (a) shows the geographical distribution of the RIVERi variable, measuring the (weighted) number of rivers flowing through province i; panel (b) shows the provinces belonging to the Alpine region, located on the left (northern) bank of the Po river. Figure 4 View largeDownload slide Water endowment. (a) Rivers and (b) Alpine region. Panel (a) shows the geographical distribution of the RIVERi variable, measuring the (weighted) number of rivers flowing through province i; panel (b) shows the provinces belonging to the Alpine region, located on the left (northern) bank of the Po river. 3.1.2. Literacy Data on literacy rates (share of individuals aged 6 years and above able to read and write) at the provincial level for the years 1871, 1881, 1901 and 1911 are those reported in the population censuses. Data on literacy rates for 1861 (used, as we will illustrate, as an instrument for literacy rates in 1871) are also from the population census. However, in 1861 Latium and Venetia were not part of Italy and census data are not available. To fill this gap, we proceed as follows. Estimates of literacy rates for the provinces of Venetia in 1861 are obtained by assuming a constant 1871–1861 ratio of literacy rates in the regions of Lombardy and Venetia (both part of the Habsburg Empire during 1815–1860 ca.). Similarly, the estimates of literacy rates for Latium in 1861 assume a constant 1871–1861 ratio of literacy rates of Latium on the one hand, and of the macro-area formed by Emilia, Umbria and the Marches on the other hand. The underlying assumption is that, between 1861 and 1871, literacy rates of the regions forming the Papal States (during 1815–1860 ca.) evolved similarly. Figure 5 illustrates the geographical distribution of literacy rates in 1871 and 1911. As a result of very different socioeconomic developments and pre-unitarian policies, the North-South regional divide is particularly marked (Ciccarelli and Weisdorf, 2016). In 1871, the north-western regions registered literacy rates of about 50%, central and north-eastern regions of about 30% and southern regions of about 15%. In 1911, after four decades of mandatory primary public schooling, literacy rates reached some 75% in the north-west, 55 in the north-east and in the center and about 35% in the south. Figure 5 View largeDownload slide Literacy rates. Share of population aged 6 years and above able to read and write. Figure 5 View largeDownload slide Literacy rates. Share of population aged 6 years and above able to read and write. 3.2. Market potential In order to analyze the importance of domestic market access as a key driver of the spatial distribution of economic activity, a sound measure of accessibility to demand is required. In line with Klein and Crafts (2012), we construct market potential estimates for each Italian province i between 1871 and 1911 using Harris (1954)’s formula, that is, as a weighted average of GDP (or total value added) of all provinces j2: MKTPOTit=∑j=1NGDPjt×dij−1, (1) with dij the great circle distance in kilometer between the centroids of provinces i and j.3 In practice, this indicator equates the potential demand for goods and services produced in a given location with that location’s proximity to consumer markets. Thus, it can be interpreted as the volume of economic activity to which a region has access to, after having taken into account the necessary transport costs to cover the distance to reach other provinces. An important point is that historical GDP estimates at the provincial level for the case of Italy are not available. Thus, we proxy for provincial GDP by allocating total regional GDP (NUTS-2 units) estimates for 1871, 1881, 1901 and 1911 to provinces (NUTS-3 units) using the provincial shares of regional population obtained by population census (see the Appendix for further details). Figure 6 shows that the values of Harris market potential are increasingly concentrated in north-western provinces. However, estimated market potential is also high in Florence, Rome, Naples and Palermo, that is the provinces with the pre-unitarian city capitals of, respectively, the Grand Duchy of Tuscany, the Papal States and the Kingdom of the Two Sicilies. This last evidence fits perfectly well with the intuition of Fenoaltea (2003, 1073–74) who, in his study on Italian regional industrialization, noticed that in the early 1870s ‘The industrial, manufacturing regions are those with the former capitals, of the preceding decades and centuries’ and that ‘In such a context the appropriate unit of analysis is not in fact the region, but (in Italy) the much smaller province’. Figure 6 View largeDownload slide Harris market potential. Figure 6 View largeDownload slide Harris market potential. 3.3. Expected sectoral effects of market potential and literacy The international literature, as we briefly summarized at the beginning of this section, shows that both market potential and factor endowment contribute to explain industrial location patterns in Britain, Poland, Spain and in the USA during the late 19th and early 20th centuries. However, the effects are generally sector-specific. Recall for instance that both Crafts and Mulatu (2006) for the UK and Klein and Crafts (2012) for the USA show that market potential was an attraction especially for industries with large plant size. This finding is important in that it supports the view that the effect of market potential may depend on the degree of scale economies prevailing in the various sectors. Effectively, economic geography theory predicts that economic activities with increasing returns to scale tend to establish themselves in regions that enjoy good market access, while the location of economic activities with constant returns technologies is mainly influenced by factor endowment. Sectoral differences may also exist in the effect of regional skill endowment on industrial location. Standard neoclassical trade theory (Heckscher–Ohlin model) predicts that high-skilled labor intensive industries tend to be concentrated in regions with higher endowment of high-skilled labor. Crafts and Mulatu (2006) show that in 19th-century Britain educated workers were an important input for the chemical sector, but not for textiles. Based on these economic geography arguments, one may broadly expect that the location of sectors characterized by high and medium capital to labor ratio (K/L), inherently tied to skill intensity and increasing returns to scale, was relatively more influenced by market potential. Table 2 groups the Italian manufacturing sectors into light and heavy industries depending on their capital intensity (the Appendix reports maps illustrating the spatial distribution of selected high and low K/L sectors). The table reports in particular data on horse-power per worker (HP/L), often used to proxy capital intensity (e.g., Broadberry and Crafts, 1990). The table is mostly based on the data from the first industrial census of 1911 given in Zamagni (1978), and the 0.6 for 1911 threshold for ‘K/L’ inferred from Federico (2006).4 It also uses the disaggregated data for the engineering sector by Fenoaltea (2016). There is little argument in the literature that metalmaking, paper and chemicals represent high ‘K/L’ sectors, while non-metallic mineral products, wood, leather and clothing are low ‘K/L’ sectors. Within the textile industry, cotton and wool belong to the high ‘K/L’ group, while silk belongs to the low one. The production of raw silk, an activity at the boundary between agriculture and manufacturing, is a very labor-intensive activity, suitable for water-rich and densely populated areas, such as the North-West of Italy. In addition, silk represented a leading component of Italian exports toward North-Western Europe, and being a luxury good, beyond the reach of most of 19th-century Italians (Federico and Tena-Junguito, 2014). As far as the engineering industry is concerned, Fenoaltea (2016) shows that the average value of K/L = 0.33 for the whole industry actually includes values as low as 0.20 for blacksmiths and as high as 0.60 for shipyards, reflecting the fact that the manufacturing of major naval vessels (but also steam locomotives) was far more sophisticated from a technological point of view than the maintenance of agricultural tools by blacksmiths.5 A final note on the foodstuffs sector is that, as warned by Zamagni (1978), the relatively high level of horse-power per worker should not be misinterpreted, since it was largely due to the traditional flour-milling industry. Table 2 Horse-power per worker (HP/L) in 1911 High HP/L Metalmaking 2.62 Chemicals and rubber 1.30 Foodstuffsa 0.94 Textiles     Cotton 0.85     Wool 0.78 Paper 0.73 Engineering     Machinery 0.61     Shipbuilding 0.60 Low HP/L Non-metallic mineral products 0.36 Wood 0.23 Engineering     Blacksmiths 0.20 Textiles     Silk 0.11 Leather 0.09 Clothing 0.07 High HP/L Metalmaking 2.62 Chemicals and rubber 1.30 Foodstuffsa 0.94 Textiles     Cotton 0.85     Wool 0.78 Paper 0.73 Engineering     Machinery 0.61     Shipbuilding 0.60 Low HP/L Non-metallic mineral products 0.36 Wood 0.23 Engineering     Blacksmiths 0.20 Textiles     Silk 0.11 Leather 0.09 Clothing 0.07 aFoodstuffs is net of sugar (with K/L = 2.18). Source: Zamagni (1978) and Fenoaltea (2016) for machinery, shipbuilding and blacksmiths. The ‘K/L’ figure for machinery is in particular the average of rail-guided vehicles, other heavy equipment and other ordinary machinery. Table 2 Horse-power per worker (HP/L) in 1911 High HP/L Metalmaking 2.62 Chemicals and rubber 1.30 Foodstuffsa 0.94 Textiles     Cotton 0.85     Wool 0.78 Paper 0.73 Engineering     Machinery 0.61     Shipbuilding 0.60 Low HP/L Non-metallic mineral products 0.36 Wood 0.23 Engineering     Blacksmiths 0.20 Textiles     Silk 0.11 Leather 0.09 Clothing 0.07 High HP/L Metalmaking 2.62 Chemicals and rubber 1.30 Foodstuffsa 0.94 Textiles     Cotton 0.85     Wool 0.78 Paper 0.73 Engineering     Machinery 0.61     Shipbuilding 0.60 Low HP/L Non-metallic mineral products 0.36 Wood 0.23 Engineering     Blacksmiths 0.20 Textiles     Silk 0.11 Leather 0.09 Clothing 0.07 aFoodstuffs is net of sugar (with K/L = 2.18). Source: Zamagni (1978) and Fenoaltea (2016) for machinery, shipbuilding and blacksmiths. The ‘K/L’ figure for machinery is in particular the average of rail-guided vehicles, other heavy equipment and other ordinary machinery. 4. Econometric analysis This section examines the relevance of factor endowment and (domestic) market potential in shaping the location of industries across Italian provinces during the period 1871–1911. Industrial location is measured in relative terms, that is by using the log of the location quotient, ln⁡(LQ). Market potential is measured by the log of Harris (1954) formula ( ln⁡(Mktpot)). As for factor endowment, we focus on labor skills and water abundance. Labor skills are measured by the log of literacy, ln⁡(Literacy), that is, the share of people aged 6 years and above who are able to read and write. Water supply is measured by the continuous, but time-invariant, variable ln⁡(River), and its interaction with the dummy variable Alpine, indicating if the province belongs to the Alpine region. Following Combes and Gobillon (2015), we test the effect of market potential, literacy and natural advantages in two steps. In the first step, we exploit the panel structure of the data (69 provinces for four time periods) to assess the effect of time-varying variables (i.e., ln⁡(Mktpot) and ln⁡(Literacy)), while in the second step we estimate the effect of time-invariant variables (i.e., water abundance). 4.1. The effect of market potential and literacy 4.1.1. Empirical strategy In this section, we discuss the estimated effects of the two time-varying variables (i.e., ln⁡(Mktpot) and ln⁡(Literacy)). Differently from Combes and Gobillon (2015), however, in this first step we control for time-invariant unobserved heterogeneity by using a simple semiparametric model with a smooth spatial trend (the so-called Geoadditive Model) (Lee and Durbán, 2011), rather than by introducing spatial-fixed effects. More formally, for each sector k, and denoting with i and t the province and time index, the model for the first step is specified as6: ln⁡(LQi,t)=α+β1ln⁡(Literacyi,t)+β2ln⁡(Mktpoti,t)+f1(Lati)+f2(Longi)+f12(Lati,Longi)+γt+εi,tεi,t ∼ iidN(0,σε2) i=1,…,N t=1,…,T. (2) Time-fixed effects (γt) are introduced in the model to control for time-related factor biases. Moreover, the geoadditive terms, that is, the smooth effect of the latitude— f1(Lati), of the longitude— f2(Longi), and of their interaction— f12(Lati,Longi)—work as control functions (CFs) to filter the spatial trend out of the residuals, and transfer it to the mean response in a model specification. Thus, they allow to capture the shape of the spatial distribution of the dependent variable, conditional on the determinants included in the model. These CFs also isolate stochastic spatial dependence in the residuals that is spatially autocorrelated unobserved heterogeneity (see also Basile et al., 2014). Thus, they can be regarded as an alternative to individual regional dummies (spatial-fixed effects) to capture unobserved spatial heterogeneity as long as the latter is smoothly distributed over space. Regional dummies peak significantly higher and lower levels of the mean response variable. If these peaks are smoothly distributed over a two-dimensional surface (i.e., if unobserved spatial heterogeneity is spatially auto-correlated), the smooth spatial trend is able to capture them.7 We simply demonstrate the validity of these statements by estimating the two competing models without explanatory variables (see the Appendix). A complication with the estimation of model (2) is given by the presence of endogenous variables— ln⁡(Mktpoti,t) and ln⁡(Literacyi,t)—on the right-hand side (r.h.s.). As for ln⁡(Mktpoti,t), NEG models describe a process characterized by reverse causality in which market potential, by attracting firms and workers, increases production in a particular location, and this, in turn, raises its market potential. ln⁡(Literacyi,t) may also be an endogenous variable. On the one hand, the availability of literate workers may foster the concentration of industrial activities in certain regions. On the other hand, however, more industrialized regions may provide an incentive to achieve education that is generally lacking in backward areas of the country. To address these issues, we extend the REML methodology to estimate the parameters of model (2) in a 2-stage ‘CF’ approach (Blundell and Powell, 2003), that is an alternative to standard instrumental variable/two-stage least square (IV–2SLS) methods. In the first stage, each endogenous variable is regressed on a set of conformable IVs (Z), using a semiparametric model. The residuals from the first stages are then included in the original model (2) to control for the endogeneity of ln⁡Mktpoti,t and ln⁡Literacyi,t. Since the second stage contains generated regressors (i.e., the first-step residuals), a bootstrap procedure is used to compute p-values [see Basile et al. (2014) for details on the bootstrap procedure]. This procedure requires finding good instruments, that is, variables that are correlated with the endogenous explanatory variables but not with the residuals of the regression. To control for the endogeneity of market potential, we follow the main empirical literature in using a measure of centrality of the region ( Centrality=∑idij−1) (Head and Mayer, 2006), and the geographical distance from the main economic center (i.e., the distance from Milan, DistMilani) (Redding and Venables, 2004; Wolf, 2007; Klein and Crafts, 2012; Martinez-Galarraga, 2012) as IVs. To control for the endogeneity of ln⁡Literacyi,t, we use its time lag ( ln⁡Literacyi,t−10). 4.1.2. Evidence for the whole manufacturing sector For the case of the whole manufacturing activity, we report in Table 3 the estimation results of the semiparametric CF approach, along with the estimation results of a fully parametric 2SLS. Obviously, we cannot use the within-group version of the 2SLS estimator, since two important instruments (DistMilani and Centralityi) are time invariant, while the third one ( ln⁡Literacyi,t−10) would be correlated with the within-group transformed error term. Thus, in order to control for spatial heterogeneity, we include a parametric nonlinear spatial trend (i.e., Lat, Lat2, Long, Long2, Lat × Long) on the r.h.s. of the pooled 2SLS model. Table 3 Whole manufacturing. Estimation results of the parametric IV (2SLS) approach and of the semiparametric control function (CF) approach Variable 2SLS Semiparametric CF Parametric terms     Intercept 5.003 −1.833***      (0.613) (0.001)      ln⁡(Mktpot) 0.610*** 0.468***      (0.000) (0.000)      ln⁡(Literacy) 0.015 0.235**      (0.934) (0.027)     Lat −0.487      (0.238)     Lat2 0.006      (0.153)     Long 0.268      (0.456)     Long2 −0.009**      (0.027)     Lat×Long −0.001      (0.940) Non-parametric terms      f1(Lat) 7.760***      (0.000)      f2(Long) 10.259***      (0.000)      f12(Lat,Long) 20.047***      (0.000)      h1(Res1) 3.007***      (0.078)      h2(Res2) 2.933***      (0.002) Diagnostics     Weak instr.-ln(Mktpot) 21.208*** 24.461*      (0.000) (0.000)     Weak instr.-ln(Literacy) 405.489*** 210.072***      (0.000) (0.000)     Wu–Hausman 10.535*** 33.056***      (0.000) (0.000)     Sargan 0.436 3.727      (0.509) (0.155) Variable 2SLS Semiparametric CF Parametric terms     Intercept 5.003 −1.833***      (0.613) (0.001)      ln⁡(Mktpot) 0.610*** 0.468***      (0.000) (0.000)      ln⁡(Literacy) 0.015 0.235**      (0.934) (0.027)     Lat −0.487      (0.238)     Lat2 0.006      (0.153)     Long 0.268      (0.456)     Long2 −0.009**      (0.027)     Lat×Long −0.001      (0.940) Non-parametric terms      f1(Lat) 7.760***      (0.000)      f2(Long) 10.259***      (0.000)      f12(Lat,Long) 20.047***      (0.000)      h1(Res1) 3.007***      (0.078)      h2(Res2) 2.933***      (0.002) Diagnostics     Weak instr.-ln(Mktpot) 21.208*** 24.461*      (0.000) (0.000)     Weak instr.-ln(Literacy) 405.489*** 210.072***      (0.000) (0.000)     Wu–Hausman 10.535*** 33.056***      (0.000) (0.000)     Sargan 0.436 3.727      (0.509) (0.155) Notes: Coefficients, e.d.f. and bootstrap p-values (in parenthesis). Time-fixed effects are included in both models. Number of observations: 276. Table 3 Whole manufacturing. Estimation results of the parametric IV (2SLS) approach and of the semiparametric control function (CF) approach Variable 2SLS Semiparametric CF Parametric terms     Intercept 5.003 −1.833***      (0.613) (0.001)      ln⁡(Mktpot) 0.610*** 0.468***      (0.000) (0.000)      ln⁡(Literacy) 0.015 0.235**      (0.934) (0.027)     Lat −0.487      (0.238)     Lat2 0.006      (0.153)     Long 0.268      (0.456)     Long2 −0.009**      (0.027)     Lat×Long −0.001      (0.940) Non-parametric terms      f1(Lat) 7.760***      (0.000)      f2(Long) 10.259***      (0.000)      f12(Lat,Long) 20.047***      (0.000)      h1(Res1) 3.007***      (0.078)      h2(Res2) 2.933***      (0.002) Diagnostics     Weak instr.-ln(Mktpot) 21.208*** 24.461*      (0.000) (0.000)     Weak instr.-ln(Literacy) 405.489*** 210.072***      (0.000) (0.000)     Wu–Hausman 10.535*** 33.056***      (0.000) (0.000)     Sargan 0.436 3.727      (0.509) (0.155) Variable 2SLS Semiparametric CF Parametric terms     Intercept 5.003 −1.833***      (0.613) (0.001)      ln⁡(Mktpot) 0.610*** 0.468***      (0.000) (0.000)      ln⁡(Literacy) 0.015 0.235**      (0.934) (0.027)     Lat −0.487      (0.238)     Lat2 0.006      (0.153)     Long 0.268      (0.456)     Long2 −0.009**      (0.027)     Lat×Long −0.001      (0.940) Non-parametric terms      f1(Lat) 7.760***      (0.000)      f2(Long) 10.259***      (0.000)      f12(Lat,Long) 20.047***      (0.000)      h1(Res1) 3.007***      (0.078)      h2(Res2) 2.933***      (0.002) Diagnostics     Weak instr.-ln(Mktpot) 21.208*** 24.461*      (0.000) (0.000)     Weak instr.-ln(Literacy) 405.489*** 210.072***      (0.000) (0.000)     Wu–Hausman 10.535*** 33.056***      (0.000) (0.000)     Sargan 0.436 3.727      (0.509) (0.155) Notes: Coefficients, e.d.f. and bootstrap p-values (in parenthesis). Time-fixed effects are included in both models. Number of observations: 276. All in all, the diagnostic tests of the 2SLS model provide encouraging evidence in favor of the chosen set of instruments. First, the Wu–Hausman test confirms that ln⁡Literacyi,t and ln⁡Mktpoti,t are endogenous. Second, the weak instrument tests confirm that the IVs are strongly correlated to the endogenous variables. Third, the Sargan test of overidentifying restrictions suggests that the DistMilani, Centralityi, and ln⁡Literacyi,t−10 are valid instruments, that is, they are uncorrelated with the error term, and thus they are correctly excluded from the estimated equation. Nevertheless, the estimated parametric 2SLS model does not properly control for unobserved spatial heterogeneity. From Table 3 it emerges indeed that the spatial variables (Lat, Lat2, Long, Long2, Lat × Long) are weakly significant. The spatial trend in the data (i.e., the smooth spatial heterogeneity) is much better captured by the semiparametric geoadditive model as indicated by the high significance of the smooth terms f1, f2, and f12 (in column semiparametric CF).8 The smooth functions of the residuals from the two first stages ( h1(Res1) and h2(Res2)) work as CFs to correct the estimated parameters for the endogeneity bias. The statistical significance of these smooth terms can also be used as endogeneity tests. In Table 3 we also report a test of the joint significance of h1(Res1) and h2(Res2), similar to the Wu–Hausman test in the parametric 2SLS approach, as well as two weak-instrument tests using the results of the semiparametric first stages (which confirm that the excluded instruments are strongly correlated to the endogenous variables, thus rejecting the hypothesis of weak instruments). Unfortunately, there is not a well-known and widely accepted test for the validity of the conditional mean restrictions imposed by the CF approach (overidentification test). A practically feasible way of testing such restrictions consists of including the excluded instruments in the CF estimate, and testing their significance. They should not be significant since the CF should pick up all of the correlation between the structural error term and (X, Z). In our case, all the external instruments turned out to be strictly exogenous. The CF estimation results for the whole manufacturing industry show that, after controlling for unobserved heterogeneity and after correcting for the endogeneity bias, the variables ln⁡(Literacy) and ln⁡(Mktpot) enter the model significantly and positively, thus proving to be important drivers of industrial location during the sample period. In particular, the positive effect of market potential corroborates the hypothesis that even during the early stage of Italy’s industrialization firms tended to settle in regions with the highest market potential (i.e., in the North West). Moreover, the elasticity of Mktpot (0.47) is much higher than that of Literacy (0.24). 4.1.3. The heterogeneous effect of market potential and literacy across sectors Table 4 looks within manufacturing and provides the estimation results of model (2) for each industrial sector. Again, the model includes a smooth spatial trend to control for unobserved spatial heterogeneity, time dummies to control for unobserved time heterogeneity and smooth terms of first-stage residuals as CFs to correct the inconsistency due to the endogeneity of ln⁡(Mktpot) and ln⁡(Literacy).9 Table 4 Marginal effects of ln(Mktpot) and ln(Literacy) Sectors Market potential Literacy Whole manufacturing 0.468*** 0.235**      (0.000) (0.027) High HP/L sectors     Metalmaking 1.502*** 1.976***      (0.000) (0.002)     Chemicals and rubber 0.653*** 0.670**      (0.000) (0.044)     Foodstuffs −0.286*** 0.272***      (0.000) (0.002)     Cotton 2.815*** −0.244      (0.000) (0.817)     Wool 0.775* −1.475*      (0.073) (0.058)     Other natural fibers 0.614** 0.220      (0.018) (0.701)     Paper 0.969*** 0.320      (0.000) (0.217)     Machinery 0.687*** 0.105      (0.000) (0.629)     Shipbuilding 0.475 2.659**      (0.469) (0.015) Low HP/L sectors     Non-metallic mineral products 0.023 −0.770***      (0.862) (0.001)     Wood −0.191*** 0.199**      (0.001) (0.019)     Blacksmith −0.222*** −0.184*      (0.000) (0.054)     Silk 2.199*** −1.035      (0.000) (0.321)     Leather −0.453*** −0.005      (0.000) (0.963)     Clothing 0.017 0.126      (0.876) (0.406) Sectors Market potential Literacy Whole manufacturing 0.468*** 0.235**      (0.000) (0.027) High HP/L sectors     Metalmaking 1.502*** 1.976***      (0.000) (0.002)     Chemicals and rubber 0.653*** 0.670**      (0.000) (0.044)     Foodstuffs −0.286*** 0.272***      (0.000) (0.002)     Cotton 2.815*** −0.244      (0.000) (0.817)     Wool 0.775* −1.475*      (0.073) (0.058)     Other natural fibers 0.614** 0.220      (0.018) (0.701)     Paper 0.969*** 0.320      (0.000) (0.217)     Machinery 0.687*** 0.105      (0.000) (0.629)     Shipbuilding 0.475 2.659**      (0.469) (0.015) Low HP/L sectors     Non-metallic mineral products 0.023 −0.770***      (0.862) (0.001)     Wood −0.191*** 0.199**      (0.001) (0.019)     Blacksmith −0.222*** −0.184*      (0.000) (0.054)     Silk 2.199*** −1.035      (0.000) (0.321)     Leather −0.453*** −0.005      (0.000) (0.963)     Clothing 0.017 0.126      (0.876) (0.406) Notes: Coefficients and bootstrap p-values (in parenthesis). The number of observations for each sector is 276 (69 province by four time points). Table 4 Marginal effects of ln(Mktpot) and ln(Literacy) Sectors Market potential Literacy Whole manufacturing 0.468*** 0.235**      (0.000) (0.027) High HP/L sectors     Metalmaking 1.502*** 1.976***      (0.000) (0.002)     Chemicals and rubber 0.653*** 0.670**      (0.000) (0.044)     Foodstuffs −0.286*** 0.272***      (0.000) (0.002)     Cotton 2.815*** −0.244      (0.000) (0.817)     Wool 0.775* −1.475*      (0.073) (0.058)     Other natural fibers 0.614** 0.220      (0.018) (0.701)     Paper 0.969*** 0.320      (0.000) (0.217)     Machinery 0.687*** 0.105      (0.000) (0.629)     Shipbuilding 0.475 2.659**      (0.469) (0.015) Low HP/L sectors     Non-metallic mineral products 0.023 −0.770***      (0.862) (0.001)     Wood −0.191*** 0.199**      (0.001) (0.019)     Blacksmith −0.222*** −0.184*      (0.000) (0.054)     Silk 2.199*** −1.035      (0.000) (0.321)     Leather −0.453*** −0.005      (0.000) (0.963)     Clothing 0.017 0.126      (0.876) (0.406) Sectors Market potential Literacy Whole manufacturing 0.468*** 0.235**      (0.000) (0.027) High HP/L sectors     Metalmaking 1.502*** 1.976***      (0.000) (0.002)     Chemicals and rubber 0.653*** 0.670**      (0.000) (0.044)     Foodstuffs −0.286*** 0.272***      (0.000) (0.002)     Cotton 2.815*** −0.244      (0.000) (0.817)     Wool 0.775* −1.475*      (0.073) (0.058)     Other natural fibers 0.614** 0.220      (0.018) (0.701)     Paper 0.969*** 0.320      (0.000) (0.217)     Machinery 0.687*** 0.105      (0.000) (0.629)     Shipbuilding 0.475 2.659**      (0.469) (0.015) Low HP/L sectors     Non-metallic mineral products 0.023 −0.770***      (0.862) (0.001)     Wood −0.191*** 0.199**      (0.001) (0.019)     Blacksmith −0.222*** −0.184*      (0.000) (0.054)     Silk 2.199*** −1.035      (0.000) (0.321)     Leather −0.453*** −0.005      (0.000) (0.963)     Clothing 0.017 0.126      (0.876) (0.406) Notes: Coefficients and bootstrap p-values (in parenthesis). The number of observations for each sector is 276 (69 province by four time points). The results were obtained by pooling the data over the 69 provinces and the four census years (1871, 1881, 1901 and 1911). For each parametric term, we report the estimated coefficient and the corresponding bootstrapped p-value. In line with our expectations, during the period 1871–1911 market potential turned out to be a key driver of the industrial location in the case of high HP/L industries. The coefficient associated to the variable ln⁡(Mktpot) is indeed positive and strongly significant in the case of metalmaking, chemicals, paper, machinery, cotton, wool and other natural fibers. The elasticity of this variable is particularly high in the case of cotton (2.8) and metalmaking (1.5). There are, however, two exceptions. The first is shipbuilding, for which the effect of market potential is not statistically significant. But in this case, the industrial location was obviously driven by the presence of a port, independently of the market potential of the region. The second exception is foodstuffs, for which the effect is negative and significant. Census data for 1911 inform us that some 50% and more of this industrial activity was accounted for by the first- and second-stage processing of wheat and other cereals (flour-milling industry, and the baking of bread, biscuits and pasta production). These activities need often to locate close to cities, regardless of their position in the urban hierarchy. Moreover, the mild climate of the South of Italy represented a strong comparative advantage for the production of some foodstuffs, such as pasta asciutta (literally ‘dry pasta’) traditionally desiccated through simple exposition to the air. The effect of ln⁡(Mktpot) was instead negative (or not significant) in the case of low HP/L industries. Silk represents an isolated, yet important, exception. It was a labor intensive sector for which the estimated effect of market potential is sizeable (2.2), and highly significant. Recall, however, that silk production was increasingly concentrated in northern provinces and largely exported to northern European countries. Thus, it is likely that proximity to northern Europe and decreasing international physical transport costs associated with railway development, more than domestic market potential, were at the heart of the relocation of the silk industry to the northern Italian provinces.10 These findings are consistent with the predictions of the Core–Periphery NEG model (Krugman, 1991), according to which economic activities with increasing returns to scale tend to establish themselves in regions that enjoy good market access. A region with a larger market potential is also characterized by a more generous compensation of local factors, while small regions that are far from the large market will have lower local wages compared to regions close to the industrial core. Thus, if everything else is unchanged, and if the firms all achieve constant returns to scale (as in the case of low HP/L industries), the increase in the price of local factors (labor, land and so on) will reduce the profitability of all the firms at that location. Within high HP/L sectors, shipbuildings, metalmaking, foodstuffs and chemicals proved to be knowledge-intensive sectors requiring educated labor force, mainly located in the north-western regions. While, for most traditional low HP/L sectors, often tied to agriculture, literacy has a negative impact or is not significant, with wood representing the only exception. 4.2. The effect of water endowment 4.2.1. Empirical strategy and evidence for the whole manufacturing sector Similarly to Combes and Gobillon (2015), in the second step of our empirical strategy we assess the effect of time-invariant variables (i.e., Alpine, ln River and their interaction). However, instead of regressing the estimated-fixed effects, αî, on Alpinei, ln⁡(Riveri) and Alpinei×ln⁡(Riveri), we regress the estimated values of the spatial trend ( fspt̂i) on these three variables. That is, we estimate the following interaction model using OLS: fspt̂i=α+β1Alpinei+β2ln⁡(Riveri)++β3Alpinei×ln⁡(Riveri)+εiεi∼iidN(0,σε2)i=1,…,Nfspt̂i=f1̂(Lati)+f2̂(Longi)+f12̂(Lati,Longi). (3) As expected, however, the OLS residuals turned out to be spatially autocorrelated. For the case of the whole manufacturing sector, the standardized Moran I test statistic on the residuals is equal to 4.539 and its p-value is 0.000. In fact, Alpinei, ln⁡(Riveri) and Alpinei×ln⁡(Riveri) capture only a portion of the large spatial heterogeneity which characterizes the spatial distribution of the location quotient net of the effect of market potential and literacy, that is, the spatial distribution of fspt̂i (the adjusted R2 of the model is 0.31). Heterogeneity among provinces, induced by an uneven distribution of immobile resources (such as natural harbors) and amenities (climate) may also be at the origin of a variety of comparative advantages. This unobserved heterogeneity may also be spatially correlated, thus introducing spatial autocorrelation in the residuals. In order to control for spatial autocorrelation, we use a semiparametric spatial filter (Tiefelsdorf and Griffith, 2007). As is well known, spatial filtering uses a set of spatial proxy variables, which are extracted as eigenvectors from the spatial weights matrix, and implants these vectors as control variables into the model. These control variables identify and isolate the stochastic spatial dependencies among observations, thus allowing model building to proceed as if the observations were independent. Specifically, we used the function SpatialFiltering from the R library spdep. This function selects eigenvectors in a semiparametric spatial filtering approach (as proposed by Tiefelsdorf and Griffith, 2007) to remove spatial dependence from linear models. The optimal subset of eigenvectors is identified by an objective function that minimizes spatial autocorrelation, that is, by finding the single eigenvector which reduces the standard variate of Moran’s I for regression residuals most, and continuing until no candidate eigenvector reduces the value by more than a tolerance value. This subset of eigenvectors is a proxy either for those spatially autocorrelated exogenous factors that have not been incorporated into a model, or for an underlying spatial process that ties the observations together. Furthermore, incorporation of all relevant eigenvectors into a model should leave the remaining residual component spatially uncorrelated. Consequently, standard statistical modeling and estimation techniques as well as interpretations can be employed for spatially filtered models. The spatially filtered OLS estimates of model (3) confirm the widely accepted idea that, over the analyzed period, the whole manufacturing activity was mainly located in proximity to Alpine rivers. Figure 7 displays in panel (a) the changes in the marginal effect of Alpine conditional on the values of ln⁡River, that is ∂fspt̂̂∂Alpine=β1+β3ln⁡River, while it shows in panel (b) the changes in the marginal effect of the continuous variable ln⁡River conditional on the values of the dummy variable Alpine, that is ∂fspt̂̂∂ln⁡River=β2+β3Alpine. Figure 7 View largeDownload slide Whole manufacturing. Marginal effect of ln(River) and Alpine with simulated 95% confidence intervals. (a) Estimated coefficient of Alpine by ln(River) and (b) Estimated coefficient of ln(River) by Alpine. Figure 7 View largeDownload slide Whole manufacturing. Marginal effect of ln(River) and Alpine with simulated 95% confidence intervals. (a) Estimated coefficient of Alpine by ln(River) and (b) Estimated coefficient of ln(River) by Alpine. The two plots, created using the R function interplot, also include simulated 95% pointwise confidence intervals (obtained using the simulation function from the R package arm of Gelman and Hill, 2006) around these marginal effects. The plot in panel (a) shows that, with increasing importance of the river (along the horizontal axis), the magnitude of the marginal effect of Alpine on the location of the whole manufacturing sector also increases (along the vertical axis). The ‘dot-and-whisker’ plot in panel (b) shows the effect of ln⁡River (measuring the importance of the river) conditional on the dummy variable Alpine. On the one hand, when Alpine = 1 (i.e., when the province belongs to the Alpine region), the simulated 95% confidence interval around the marginal effect of ln⁡River is above (and does not contain) the horizontal zero line. On the other hand, when Alpine = 0 (i.e., when the province does not belong to the Alpine region), the simulated 95% confidence interval around the marginal effect of ln⁡River is below (and does not contain) the horizontal zero line. Thus, for manufacturing, there is a positive effect of Alpine rivers, as widely suggested by previous qualitative studies of historical nature stressing the importance of water as a source of motive power and, more generally, to sustain manufacturing activities that required water throughout the year. 4.2.2. The heterogeneous effect of water endowment across sectors Table 5 shows that, once the effect of market potential and literacy is accounted for, the marginal effect of ln⁡River when Alpine = 1 is positive and statistically significant for 6 out of 15 industries. They are evenly shared between low HP/L industries (clothing, silk and leather) and high HP/L sector (chemicals, foodstuffs and other natural fibers). In the case of chemical products, we find in particular the same result holding for the whole manufacturing industry (i.e., a positive marginal effect of Alpine rivers and a negative effect of Apennine rivers). The estimated effect is instead negative or insignificant for the remaining industries. Apennine rivers have instead a positive effect in the cases of wool, silk, blacksmith and leather. Table 5 Marginal effect of ln(River) conditional on Alpine Sectors Alpine = 0 Lower bound Upper bound Alpine = 1 Lower bound Upper bound Manufacturing −0.208* −0.281 −0.133 0.127* 0.013 0.241 High HP/L sectors     Metalmaking −0.294* −0.532 −0.059 −0.283 −0.632 0.074     Chemicals and rubber −0.198* −0.333 −0.064 0.430* 0.221 0.646     Foodstuffs 0.055 −0.018 0.130 0.150* 0.041 0.257     Cotton −0.200 −0.459 0.049 −1.023* −1.389 −0.656     Wool 0.712* 0.256 1.162 −0.002 −0.679 0.681     Other natural fibers 0.029 −0.130 0.189 0.528* 0.292 0.767     Paper −0.095 −0.215 0.025 −0.225* −0.403 −0.052     Machinery −0.040 −0.098 0.015 0.031 −0.054 0.114     Shipbuilding −1.460* −2.254 −0.661 −0.557 −1.730 0.643 Low HP/L sectors     Non-metallic mineral products 0.045 −0.074 0.164 −0.397* −0.589 −0.204     Wood −0.026 −0.091 0.039 0.051 −0.051 0.152     Blacksmith 0.089* 0.046 0.132 −0.033 −0.101 0.035     Silk 0.492* 0.016 0.961 0.731* 0.034 1.452     Leather 0.075* 0.009 0.141 0.156* 0.062 0.250     Clothing 0.078 −0.036 0.188 0.189* 0.018 0.361 Sectors Alpine = 0 Lower bound Upper bound Alpine = 1 Lower bound Upper bound Manufacturing −0.208* −0.281 −0.133 0.127* 0.013 0.241 High HP/L sectors     Metalmaking −0.294* −0.532 −0.059 −0.283 −0.632 0.074     Chemicals and rubber −0.198* −0.333 −0.064 0.430* 0.221 0.646     Foodstuffs 0.055 −0.018 0.130 0.150* 0.041 0.257     Cotton −0.200 −0.459 0.049 −1.023* −1.389 −0.656     Wool 0.712* 0.256 1.162 −0.002 −0.679 0.681     Other natural fibers 0.029 −0.130 0.189 0.528* 0.292 0.767     Paper −0.095 −0.215 0.025 −0.225* −0.403 −0.052     Machinery −0.040 −0.098 0.015 0.031 −0.054 0.114     Shipbuilding −1.460* −2.254 −0.661 −0.557 −1.730 0.643 Low HP/L sectors     Non-metallic mineral products 0.045 −0.074 0.164 −0.397* −0.589 −0.204     Wood −0.026 −0.091 0.039 0.051 −0.051 0.152     Blacksmith 0.089* 0.046 0.132 −0.033 −0.101 0.035     Silk 0.492* 0.016 0.961 0.731* 0.034 1.452     Leather 0.075* 0.009 0.141 0.156* 0.062 0.250     Clothing 0.078 −0.036 0.188 0.189* 0.018 0.361 Notes: Coefficients and bootstrap confidence intervals. The number of observations for each sector is 69. Table 5 Marginal effect of ln(River) conditional on Alpine Sectors Alpine = 0 Lower bound Upper bound Alpine = 1 Lower bound Upper bound Manufacturing −0.208* −0.281 −0.133 0.127* 0.013 0.241 High HP/L sectors     Metalmaking −0.294* −0.532 −0.059 −0.283 −0.632 0.074     Chemicals and rubber −0.198* −0.333 −0.064 0.430* 0.221 0.646     Foodstuffs 0.055 −0.018 0.130 0.150* 0.041 0.257     Cotton −0.200 −0.459 0.049 −1.023* −1.389 −0.656     Wool 0.712* 0.256 1.162 −0.002 −0.679 0.681     Other natural fibers 0.029 −0.130 0.189 0.528* 0.292 0.767     Paper −0.095 −0.215 0.025 −0.225* −0.403 −0.052     Machinery −0.040 −0.098 0.015 0.031 −0.054 0.114     Shipbuilding −1.460* −2.254 −0.661 −0.557 −1.730 0.643 Low HP/L sectors     Non-metallic mineral products 0.045 −0.074 0.164 −0.397* −0.589 −0.204     Wood −0.026 −0.091 0.039 0.051 −0.051 0.152     Blacksmith 0.089* 0.046 0.132 −0.033 −0.101 0.035     Silk 0.492* 0.016 0.961 0.731* 0.034 1.452     Leather 0.075* 0.009 0.141 0.156* 0.062 0.250     Clothing 0.078 −0.036 0.188 0.189* 0.018 0.361 Sectors Alpine = 0 Lower bound Upper bound Alpine = 1 Lower bound Upper bound Manufacturing −0.208* −0.281 −0.133 0.127* 0.013 0.241 High HP/L sectors     Metalmaking −0.294* −0.532 −0.059 −0.283 −0.632 0.074     Chemicals and rubber −0.198* −0.333 −0.064 0.430* 0.221 0.646     Foodstuffs 0.055 −0.018 0.130 0.150* 0.041 0.257     Cotton −0.200 −0.459 0.049 −1.023* −1.389 −0.656     Wool 0.712* 0.256 1.162 −0.002 −0.679 0.681     Other natural fibers 0.029 −0.130 0.189 0.528* 0.292 0.767     Paper −0.095 −0.215 0.025 −0.225* −0.403 −0.052     Machinery −0.040 −0.098 0.015 0.031 −0.054 0.114     Shipbuilding −1.460* −2.254 −0.661 −0.557 −1.730 0.643 Low HP/L sectors     Non-metallic mineral products 0.045 −0.074 0.164 −0.397* −0.589 −0.204     Wood −0.026 −0.091 0.039 0.051 −0.051 0.152     Blacksmith 0.089* 0.046 0.132 −0.033 −0.101 0.035     Silk 0.492* 0.016 0.961 0.731* 0.034 1.452     Leather 0.075* 0.009 0.141 0.156* 0.062 0.250     Clothing 0.078 −0.036 0.188 0.189* 0.018 0.361 Notes: Coefficients and bootstrap confidence intervals. The number of observations for each sector is 69. Overall, the results shed a new light on the interpretation of the much debated role of water endowment on industrial location in 19th-century Italy. The historical literature stressed that during the early development of the Italian manufacturing industry, Alpine regions had a comparative advantage (over central and southern ones) stemming from water endowment. Effectively, even when the effect of market potential and literacy is duly accounted for, there is still room left for water endowment as a driver of industrial location. However, this role appears to be inherently tied to the nature of the industrial sector considered, and cannot simply be attributed to a generic Alpine/Apennine divide. Consider the case of the silk and leather industries. The descriptive analysis (reported in the Appendix) shows that silk was increasingly agglomerated in the Alpine region, while leather was spatially diffused in the South. Both industries made an intensive use of fresh water, not just as a source of motive power, but also and more generally in the various steps of their production process. For both industries, the empirical results of Table 5 show effectively that the proximity to a river was relevant independently of its position (either Alpine or Apennine). 5. Conclusions Using a new set of provincial (NUTS-3 units) data on industrial value added at 1911 prices, this paper analyzed the spatial location patterns of manufacturing activity in Italy during the period 1871–1911. Specifically, we tested the relative effect of tangible factors, such as market size and factor endowment (water abundance and literacy), as the main drivers of industrial location, ruling out the role of intangible factors such as knowledge spillovers. The results show that, after the political and economic unification of the country in 1861, Italian provinces became more and more specialized, and manufacturing activity became increasingly concentrated in a few provinces, mostly belonging to the North-West part of the country. The estimation results corroborate the hypothesis that both comparative advantages (water endowment effect) and market potential (home-market effect) have been responsible for this process of spatial concentration. These findings are in line with those available in the literature for other countries (see, e.g., Rosés (2003) for 19th-century Spain). The capital/labor (‘K/L’) ratio, related to both skill-intensity and increasing-returns-to scale, provided useful guidance in the analysis of sectoral developments. The role of market potential as a driver of industrial location emerged clearly for medium-high ‘K/L’ sectors mostly tied to the production of durable and capital goods (metalmaking, chemicals, machinery and paper), but also of consumer goods (cotton and wool). In addition, the location of metalmaking and chemicals was positively related to the availability of more educated labor force. Furthermore, once the effect of market potential and literacy is accounted for, we find evidence that the location of some traditional industries characterized by a low K/L ratio (such as silk and leather) was mainly driven by water endowment. There are of course more historical industrial location developments in heaven and earth than in our theoretical and empirical models. The case of shipbuilding, where only literacy emerged as an important driver of industrial location, does not fit well for instance within our interpretive scheme. With these caveats in mind, our results confirm the importance of using data that are disaggregated at the sectoral level when investigating the main factors influencing regional industrial location. Our findings are in line with the Core–Periphery NEG model (Krugman, 1991) that predicts increasing polarization and regional specialization as a result of economic integration. In this model, agglomeration economies derive from the interaction among economies of scale, transportation costs and market size, while intangible external economies (such as information spillovers) do not play any role. Today NEG models are strongly criticized on the basis of the observation that they focus on forces and processes that were important a century ago but are much less relevant today. The NEG approach seems less and less applicable to the actual location patterns of advanced economies. Nevertheless, the current economic geography of fast-growing countries like Brazil, China and India is highly reminiscent of the economic geography of Western countries during the 19th century, and it fits well into the NEG framework. Historical analyses like the one presented in this paper might thus help our understanding of the current process of economic modernization of developing countries. Footnotes 1 The increasing domestic market integration (decreasing trade costs) during the sample period can be documented through several evidences. First, in the early 1860s internal trade barriers were eliminated, and the mild external trade tariff previously adopted by the Kingdom of Savoy was extended to the rest of the country (Ciccarelli and Proietti, 2013). Second, internal transport costs were reduced through the development of the railway network: 8 km in 1839, 2400 km in 1860, 9100 km in 1880, and 15,300 km in 1910, when the network was essentially completed (ISTAT, 1958, 137). Finally, the reduction in shipping and railway rates during 1870–1910 is documented in Missiaia (201648). 2 Harris (1954)’s formula can be derived from a NEG theoretical model (Combes et al., 2008). An alternative way of obtaining a structural estimate of market potential based on NEG models was proposed by Redding and Venables (2004). The latter, however, requires data on bilateral trade flows that are not available at regional level for our sample period. 3 In order to measure Harris’ market potential, we rely on geodesic distances between locations. Recent historical studies (see e.g., Martínez-Galarraga, 2014) consider more sophisticated measures of bilateral transport costs. These measures take into consideration different transport modes, the routes used in the transportation of commodities by mode and the respective freight rates; they also take into account that their evolution over time can vary for a number of reasons. The adoption of this approach is beyond the scope of the present paper, since it requires a large amount of historical information that is not available using current data sources. Rather, it will be considered as an objective of our future research agenda. 4 Federico (2006) distinguishes between light industries and heavy industries. The former are those that ‘are featured by a (relatively) low capital intensity and by the prevailing orientation toward the final consumer’, the latter ‘are more capital-intensive and produce mainly inputs for other sectors’. The author (see in particular Table 2.6 therein), uses the industrial census of 1927 and considers as heavy industries those with a ‘K/L’ value above 0.7. Our data refer to 1911, and a 0.6 threshold for ‘K/L’ dovetails better with the data at hand. 5 The historical data and methodology used in Zamagni (1978) and Fenoaltea (2016) for the engineering sector differ, so that the average ‘K/L’ value for the year 1911 proposed by Zamagni (1978) (‘K/L’ = 0.51) and by Fenoaltea (2016) (‘K/L’ = 0.33) are difficult to compare. 6 This specification is different from the one used in the related literature (e.g., by Ellison and Glaeser, 1999; Midelfart-Knarvik et al., 2000; Wolf, 2007). These authors pool the data by regions, sectors and time, and regress the location quotient on a set of interactions between the vectors of location characteristics (factor endowment and market potential) and a vector of industry characteristics (measuring industries’ factor intensities and the share of intermediate inputs in GDP). Our dataset is rich enough to avoid considering the above interaction and pooling of the data, and it allows us to estimate separate models for each industrial sector. 7 The smooth components of the spatial trend in equation 2—f1, f2 and f12—are approximated by tensor product smoothers (Wood, 2006). As is well known, any semiparametric model can be expressed as a mixed model and, thus, it is possible to estimate all the parameters using restricted maximum-likelihood methods (REMLs). To estimate this model, we used the method described by Wood (2006) which allows for automatic and integrated smoothing parameters selection through the minimization of the REML. Wood has implemented this approach in the R package mgcv. 8 For the nonparametric smooth terms (i.e., the spatial trend and the first-stage residuals), Table 3 also shows the estimated degree of freedom (e.d.f.), a broad measure of nonlinearity (an e.d.f. equal to 1 indicates linearity, while a value higher than 1 indicates nonlinearity). 9 The results for these controls are not reported in Table 4, but are available upon request. 10 Up to the late 1880s, when Italy started a nonsense trade-war with France, nearly half of Italy’s total exports, albeit rather limited in size, were directed to the French market (Federico et al., 2011, 42–43). Acknowledgements We are grateful to Brian A'Hearn, Stefano Fachin, and Jacob Weisdorf for useful comments and suggestions. Research presented at the Berlin Colloquium of Economic History (Berlin, 2015), the Jean Paelinck Seminar of Spatial Econometric (Dijon 2015), the annual conferences of the European Economic History Society (EHES 2015, Pisa), Economic History Society (EHS 2015, Telford), the Italian Association of Regional Science (AISRe 2015, Rende), and the Italian Economic Association (SIE 2015, Naples). References A’Hearn B. , Venables A. ( 2013 ) Regional disparities: internal geography and external trade. In Toniolo G. (ed.) The Oxford Handbook of the Italian Economy since Unification . Oxford : Oxford University Press . Arbia G. , Dedominicis L. , De Groot H. ( 2013 ) The spatial distribution of economic activities in Italy. A sectorally detailed spatial data analysis using local labour market areas . Regional Studies , 47 : 405 – 418 . Google Scholar CrossRef Search ADS Basile R. , Durbán M. , Mínguez R. , Montero J. M. , Mur J. ( 2014 ) Modeling regional economic dynamics: spatial dependence, spatial heterogeneity and nonlinearities . Journal of Economic Dynamics and Control , 48 : 229 – 245 . Google Scholar CrossRef Search ADS Blundell R. , Powell J. ( 2003 ) Endogeneity in nonparametric and semiparametric regression models. In Dewatripont M. , Hansen L. , Turnsovsky S. J. (eds) Advances in Economics and Econometrics Theory and Application , pp. 312–357. Cambridge : Cambridge University Press . Broadberry S. N. , Crafts N. F. ( 1990 ) Explaining Anglo-American productivity differences in the mid-twentieth century . Oxford Bulletin of Economics and Statistics , 52 : 375 – 402 . Google Scholar CrossRef Search ADS Cafagna L. ( 1989 ) Dualismo e Sviluppo nella Storia d’Italia . Venice : Marsilio . Ciccarelli C. , Fenoaltea S. ( 2009 ) La Produzione Industriale delle Regioni d’Italia, 1861–1913: una Ricostruzione Quantitativa. 1. Le Industrie Non Manifatturiere . Rome : Bank of Italy . Ciccarelli C. , Fenoaltea S. ( 2013 ) Through the magnifying glass: provincial aspects of industrial growth in post-unification Italy . Economic History Review , 66 : 57 – 85 . Google Scholar CrossRef Search ADS Ciccarelli C. , Fenoaltea S. ( 2014 ) La Produzione Industriale delle Regioni d’Italia, 1861–1913: una Ricostruzione Quantitativa. 2. Le Industrie Estrattivo-Manifatturiere . Rome : Bank of Italy . Ciccarelli C. , Proietti T. ( 2013 ) Patterns of industrial specialisation in post-unification Italy . Scandinavian Economic History Review , 61 : 259 – 286 . Google Scholar CrossRef Search ADS Ciccarelli C. , Weisdorf J. L. ( 2016 ) The effect of the Italian unification on the comparative regional development in literacy, 1821–1911 . WP CEIS No. 392, University of Rome Tor Vergata . Combes P.-P. , Gobillon L. ( 2015 ) The empirics of agglomeration economies. In Gilles Duranton , Vernon Henderson J. , Strange William C. (eds) Handbook of Regional and Urban Economics SET , pp. 247 – 348 . Amsterdam : Elsevier . Combes P.-P. , Mayer T. , Thisse J.-F. ( 2008 ) Economic Geography: The Integration of Regions and Nations . Princeton, NJ : Princeton University Press . Crafts N. , Mulatu A. ( 2005 ) What explains the location of industry in Britain, 1871–1931? Journal of Economic Geography , 5 : 499 – 518 . Google Scholar CrossRef Search ADS Crafts N. , Mulatu A. ( 2006 ) How did the location of industry respond to falling transport costs in Britain before World War I? Journal of Economic History , 66 : 575 – 607 . Google Scholar CrossRef Search ADS Ellison G. , Glaeser E. L. ( 1999 ) The geographic concentration of industry: does natural advantage explain agglomeration? American Economic Review , 89 : 311 – 316 . Google Scholar CrossRef Search ADS Federico G. ( 2006 ) Industrial structure (1911–2001). In Renato Giannetti, , Michelangelo Vasta (eds) Evolution of Italian Enterprises in the 20th Century , pp. 15 – 47 . Heidelberg-New York : Springer . Google Scholar CrossRef Search ADS Federico G. , Natoli S. , Tattara G. , Vasta M. ( 2011 ) Il commercio estero italiano: 1862–1950, Collana storica della Banca d’Italia/Serie “Statistiche storiche”: Collana storica della Banca d’Italia, Laterza. Federico G. , Tena-Junguito A. ( 2014 ) The ripples of the industrial revolution: exports, economic growth, and regional integration in Italy in the early nineteenth century . European Review of Economic History , 18 : 349 – 369 . Google Scholar CrossRef Search ADS Felice E. ( 2013 ) Regional income inequality in Italy in the long run (1871–2001). Patterns and determinants. UHE Working Paper, Universitat Autonoma de Barcelona. Fenoaltea S. ( 2003 ) Peeking backward: regional aspects of industrial growth in post-unification Italy . Journal of Economic History , 63 : 1059 – 1102 . Fenoaltea S. ( 2004 ) Textile production in Italy’s regions, 1861–1913 . Rivista di Storia Economica , 20 : 145 – 174 . Fenoaltea S. ( 2011 ) The Reinterpretation of Italian Economic History: from Unification to the Great War . New York : Cambridge University Press . Google Scholar CrossRef Search ADS Fenoaltea S. ( 2016 ) The measurement of production: lessons from the engineering industry in Italy, 1911 . Research in Economic History , 32 : 73 – 145 . Google Scholar CrossRef Search ADS Fujita M. , Krugman P. , Venables A. ( 1999 ) The Spatial Economics: Cities, Regions and International Trade . Cambridge (MA) : MIT Press . Gelman A. , Hill J. ( 2006 ) Data Analysis using Regression and Multilevel/Hierarchical Models . Cambridge : Cambridge University Press . Google Scholar CrossRef Search ADS Harris C. D. ( 1954 ) The market as a factor in the localization of industry in the United States . Annals of the Association of American Geographers , 44 : 315 – 348 . Head K. , Mayer T. ( 2006 ) Regional wage and employment responses to market potential in the EU . Regional Science and Urban Economics , 36 : 573 – 594 . Google Scholar CrossRef Search ADS IGM . ( 2007 ) Classificazione dei Corsi d’acqua d’Italia. Florence: Istituto Geografico Militare. ISTAT . ( 1958 ) Sommario di Statistiche Storiche Italiane: 1861–1955 . Rome : ISTAT . Klein A. , Crafts N. ( 2012 ) Making sense of the manufacturing belt: determinants of US industrial location, 1880–1920 . Journal of Economic Geography , 12 : 775 – 807 . Google Scholar CrossRef Search ADS Krugman P. , Venables A. J. ( 1995 ) Globalization and the inequality of nations . The Quarterly Journal of Economics , 110 : 857 – 880 . Google Scholar CrossRef Search ADS Krugman P. R. ( 1991 ) Geography and Trade . Cambridge, MA : MIT Press . Lee D.-J. , Durbán M. ( 2011 ) P-spline ANOVA-type interaction models for spatio-temporal smoothing . Statistical Modelling , 11 : 49 – 69 . Google Scholar CrossRef Search ADS Martinez-Galarraga J. ( 2012 ) The determinants of industrial location in Spain, 1856–1929 . Explorations in Economic History , 49 : 255 – 275 . Google Scholar CrossRef Search ADS Martínez-Galarraga J. ( 2014 ) Market potential estimates in history: a survey of methods and an application to Spain, 1867–1930. EHES Working Papers in Economic History No. 51. Midelfart-Knarvik K. H. , Overman H. G. ( 2002 ) Delocation and European integration: is structural spending justified? Economic policy , 17 : 321 – 359 . Google Scholar CrossRef Search ADS Midelfart-Knarvik K. H. , Overman H. G. , Venables A. ( 2000 ) Comparative advantage and economic geography: estimating the determinants of industrial location in the EU . Centre for Economic Policy Research. Discussion Papers , No. 2618. Missiaia A. ( 2016 ) Where do we go from here? Market access and regional development in Italy (1871–1911) . European Review of Economic History , 20 : 215 – 241 . Google Scholar CrossRef Search ADS Puga D. ( 1999 ) The rise and fall of regional inequalities . European Economic Review , 43 : 303 – 334 . Google Scholar CrossRef Search ADS Redding S. , Venables A. J. ( 2004 ) Economic geography and international inequality . Journal of International Economics , 62 : 53 – 82 . Google Scholar CrossRef Search ADS Rosés J. R. ( 2003 ) Why isn’t the whole of Spain industrialized? New economic geography and early industrialization, 1797–1910 . Journal of Economic History , 63 : 995 – 1022 . Google Scholar CrossRef Search ADS Sundar V. J. , Ramesh R. , Rao P. S. , Saravanan P. , Sridharnath B. , Muralidharan C. ( 2001 ) Water management in leather industry . Journal of Scientific and Industrial Research , 60 : 443 – 450 . Tiefelsdorf M. , Griffith D. A. ( 2007 ) Semiparametric filtering of spatial autocorrelation: the eigenvector approach . Environment and Planning A , 39 : 1193 – 1221 . Google Scholar CrossRef Search ADS Tirado D. A. , Paluzie E. , Pons J. ( 2002 ) Economic integration and industrial location: the case of Spain before World War I . Journal of Economic Geography , 2 : 343 – 363 . Google Scholar CrossRef Search ADS Wolf N. ( 2007 ) Endowments vs. market potential: what explains the relocation of industry after the Polish reunification in 1918? Explorations in Economic History , 44 : 22 – 42 . Google Scholar CrossRef Search ADS Wood S. ( 2006 ) Generalized Additive Models: an Introduction with R . Boca Raton, Florida : CRC Press . Zamagni V. N. ( 1978 ) Industrializzazione e Squilibri Regionali in Italia: Bilancio dell’età Giolittiana . Bologna : Mulino . Appendix A.1. Map of Italy’s provinces at 1911 borders The name of each region (roughly NUTS-2 units), in bold, is followed by the name (and tag) of its provinces (roughly NUTS-3 units). Different shadows of gray simply highlight regional borders. There were 16 regions and 69 provinces. The list follows a broad North to South ordering. PIEDMONT: Alessandria (AL), Cuneo (CN), Novara (NO), Turin (TO) LIGURIA: Genoa (GE), Porto Maurizio (PM) LOMBARDY: Bergamo (BG), Brescia (BS), Como (CO), Cremona (CR), Mantua (MN), Milan (MI), Pavia (PV), Sondrio (SO) VENETIA: Belluno (BL), Padua (PD), Rovigo (RO), Treviso (TV), Udine (UD), Venice (VE), Verona (VR), Vicenza (VI) EMILIA: Bologna (BO), Ferrara (FE), Forlí (FO), Modena (MO), Parma (PR), Piacenza (PC), Ravenna(RA), Reggio Emilia (RE) TUSCANY: Arezzo (AR), Florence (FI), Grosseto (GR), Leghorn (LI), Lucca (LU), Massa Carrara (MS), Pisa (PI), Siena (SI) MARCHES: Ancona (AN), Ascoli Piceno (AP), Macerata (MC), Pesaro (PE) UMBRIA: Perugia (PG) LATIUM: Rome (RM) ABRUZZI: Aquila (AQ), Campobasso (CB), Chieti (CH), Teramo (TE) CAMPANIA: Avellino (AV), Benevento (BN), Caserta (CE), Naples (NA), Salerno (SA) APULIA: Bari (BA), Foggia (FG), Lecce (LE) BASILICATA: Potenza (PZ) CALABRIA: Catanzaro (CZ), Cosenza (CS), Reggio Calabria (RC) SICILY: Caltanissetta (CL), Catania (CT), Girgenti (AG), Messina (ME), Palermo (PA), Syracuse (SR), Trapani (TP) SARDINIA: Cagliari (CA), Sassari (SS) View largeDownload slide View largeDownload slide A.2. Data A.2.1. Value added for manufacturing sectors Provincial value added data at 1911 prices for the 12 manufacturing sectors listed in Table 1 are from Ciccarelli and Fenoaltea (2013). The data are available at http://onlinelibrary.wiley.com/doi/10/j.1468-0289.2011.00643.x/full (see in particular the Supplementary Materia). Ciccarelli and Fenoaltea obtained the provincial (i.e., NUTS 3) figures as follows. They first produced annual 1861–1913 regional (i.e., NUTS 2) value added data at 1911 prices as a result of a long-term project sponsored by the bank of Italy (Ciccarelli and Fenoaltea, 2009, 2014) (see https://www.bancaditalia.it/pubblicazioni/altre-pubblicazioni-storiche/produzione-industriale-1861-1913/). The regional estimates of industrial value added are based, whenever possible and depending on the availability of historical sources, on physical production data. To give an example, consider the chemical industry. The historical sources are extremely detailed and allow one to essentially track the 1861–1913 physical production of about 100 products (sulfuric acid, soda nitric acid, Leblanc hydrochloric acid, matches, superphosphate, Thomas slang, etc.) separately for each of the 16 regions. These physical production regional series (say, tons of output) are weighted by a unitarian value added (lire of value added per ton, or other physical unit) evaluated at the national level for the year 1911 and based on historical data on capital and wages. Regional 1861–1913 value added series at 1911 prices for the whole chemical sector are then obtained by summation of the separated regional series referring to the various products. The regional estimates thus account for differences in productivity among regions and industrial sectors. Once the regional value added data for each industrial sector (VAREGIO) were obtained, the authors allocated them to provinces (VAPROV), by using the provincial labor force shares (LFSHPROV) of regional totals, separately for each industrial sector. For each given region and time, provincial value added in province i and sector j has been obtained as: VAPROVi,j=VAREGIOj*LFSHPROVi,j. As a consequence, while annual 1861–1913 regional estimates (VAREGIO) are based on a rich set of detailed historical sources, provincial estimates (VAPROV) rest essentially on the information on the labor force as reported in the population censuses and, thus, are only available for the years 1871, 1881, 1901 and 1911 (the 1891 population census was not taken, and the first industrial census was carried out in 1911). Following this approach, we used the population censuses of 1871, 1881, 1901 and 1911 to further disaggregate at the provincial level the regional value added data for the textile and engineering industries. Detailed regional value added data for the textile industry and its sectoral components (wool, silk, cotton and other natural fibers) are reported in Fenoaltea (2004), while regional value added data for the engineering industry and its sectoral components (shipbuilding, machinery and blacksmith) are provided by Ciccarelli and Fenoaltea (2014). It is worth noticing that population censuses do not generally report information on industrial sectors, rather on individual professions. The latter, as is customary, must be mapped into industrial sectors, which always involves a certain degree of arbitrariness. The details are too numerous to be even partially illustrated here. However, just to give the basic idea, we allocated, province by province, and census year by census year, to the cotton sectors professions such as cotton spinner or cotton weaver, and to the blacksmith sector professions such as blacksmith or coppersmith. Finally, we illustrate the geographical distribution of location quotients. For the reason of space, we consider metalmaking and leather as representative of ‘high’ and ‘low’ K/L sectors, respectively. We consider instead separately the main sectoral components of the textile industry (cotton and silk). Figure A1 illustrates the case of ‘high’ K/L sectors in 1871 and 1911. There is a clear predominance of the West side on Central and Northern Italy. Among northern provinces, a marked difference between sub-Alpine provinces and those along the Po valley is evident. At the same time, however, some reallocation within the Center is also evident, with a shift away from the coastal Tyrrhenian regions (Latium and Southern Tuscany). In the South light colors prevail in 1871 and 1911, especially so along the backbone represented by the Central and Southern Apennines provinces down to Calabria, at the toe of Italy’s boot. Figure A1 View largeDownload slide Metalmaking (high K/L): choropleth maps of LQ values. (a) 1871 and (b) 1911. Figure A1 View largeDownload slide Metalmaking (high K/L): choropleth maps of LQ values. (a) 1871 and (b) 1911. Figure A2 considers ‘low’ K/L sectors in 1871 and 1911. There is little change here. In 1871, and even more in 1911, a clear North-South gradient or spatial trend is present. It is interesting to note that Figure A1 and Figure A2 essentially complement each other, suggesting possibly alternative driving forces for their industrial location. Figure A2 View largeDownload slide Leather (low K/L): choropleth maps of LQ values. (a) 1871 and (b) 1911. Figure A2 View largeDownload slide Leather (low K/L): choropleth maps of LQ values. (a) 1871 and (b) 1911. Figures A3 and A4 refer finally to the geographical distribution of cotton and silk, representing, together with metalmaking, the most agglomerated sectors in 1911. The predominant role of Alpine regions emerges clearly, with Piedmont and Lombardy alone accounting for about 75% of value added in textiles in 1913 (Fenoaltea, 2004). However, the sectoral reallocation between 1871 and 1911, with the whitening of the maps for the southern provinces of Campania and Sicily is also evident, especially for the cotton industry. These facts are largely consistent with the progressive mechanization of the cotton industry, also sustained by rising protection, with import duties increased during the late 1870s and 1880s (Fenoaltea, 2004). It is also interesting to note that silk producers were instead traditionally favorable to free trade (Cafagna, 1989; Fenoaltea, 2011). Figure A3 View largeDownload slide Cotton (high K/L): choropleth maps of LQ values. (a) 1871 and (b) 1911. Figure A3 View largeDownload slide Cotton (high K/L): choropleth maps of LQ values. (a) 1871 and (b) 1911. Figure A4 View largeDownload slide Silk (low K/L): choropleth maps of LQ values. (a) Year: 1871 and (b) Year: 1911. Figure A4 View largeDownload slide Silk (low K/L): choropleth maps of LQ values. (a) Year: 1871 and (b) Year: 1911. A.2.2. Gross domestic product Historical GDP estimates at the provincial level for the case of Italy are not available. Thus, we proxy for provincial GDP by allocating total regional GDP (NUTS-2 units) estimates for 1871, 1881, 1901 and 1911 to provinces (NUTS-3 units) using the provincial shares of regional population obtained by population censuses. Data on GDP at historical borders (NUTS-2 units) were kindly provided by E. Felice. GDP includes of course industry, agriculture and services. It is important to stress that the industrial component of the regional GDP estimates by Felice is largely based on the statistical reconstructions by Ciccarelli and Fenoaltea considered before [we refer the reader to Felice (2013) for further details]. A.3. Geoadditive model versus fixed effects model In this appendix, we compare the estimation results of the geoadditive model and the fixed effects model using as dependence variable the location quotient for the whole manufacturing sector. Starting from a model without explanatory variables, we report in Figure A5 the choropleth map of the estimated fixed effects, αî, from a simple model like ln⁡(LQi,t)=αi+εi,t along with the choropleth map of the predicted values of the spatial trend, sptî=f1̂+f2̂+f12̂,obtained from the estimation of a simple geoadditive model without explanatory variables, ln⁡(LQi,t)=α+f1(noi)+f2(ei)+f12(noi,ei)+εi,t. Figure A5 View largeDownload slide Whole manufacturing. Comparing estimated-fixed effects and spatial trend. (a) Estimated-fixed effects and (b) Estimated spatial trend. Figure A5 View largeDownload slide Whole manufacturing. Comparing estimated-fixed effects and spatial trend. (a) Estimated-fixed effects and (b) Estimated spatial trend. By comparing Figure A5 and Figure 2, we may conclude that both models capture well the spatial distribution of the location quotient for the whole manufacturing activity. Extending the two models to include the two time-varying variables (the log of Harris market potential and the log of labor skills), we get the results reported in Table A1 which confirm that the two approaches can be used as alternative specifications to control for unobserved spatial heterogeneity. The sign of the coefficients estimated with the two models is the same, although the magnitude and level of significance is a bit different. However, the model with spatial trend has an important advantage over the fixed effects model. In particular, the geoadditive model allows us to save degrees of freedom. The fixed effect model uses 69 + 2 = 71 degrees of freedom (one coefficient for each of the 69 regional dummies, plus two parameters for the two explanatory variables), while the model with the spatial trend uses 45 effective degrees of freedom (e.d.f. for the spatial trend plus three parameters for the intercept and the two explanatory variables). To estimate the geoadditive model, we used a number of knots equal to 14 for each of the two univariate components of the spatial trend—f1 and f2—and seven knots for the bivariate term—f12. Thus, in total we used 14 + 14 + 49 = 77 knots, that is, a large number of knots (larger than the number of fixed effects!) to better capture nonlinearities in the spatial trend. However, the penalized spline method reduced the number of effectively used parameters (i.e., the e.d.f.) to 45 (exactly 45.123). It is worth noticing that the degree of smoothing, that is the degree of penalization, is automatically determined by the REML estimation procedure, and thus there is no arbitrary choice from the researcher. This suggests that we do not need to remove all the between-group variability in the model to filter the spatial unobserved heterogeneity. What really matters is to remove any systematic spatial pattern from the residuals which might be correlated to the explanatory variables (i.e., the source of endogeneity). We conclude that the spatial trend model must be preferred to the fixed effects model as a way to control spatial heterogeneity and to get reasonable results of the effects of the explanatory variables. Table A1 Whole manufacturing Variable Fixed effects Geoadditive model Intercept −1.290*** (0.444) Ln(Mktpot) 0.096** 0.144*** (0.048) (0.047) Ln(Literacy) −0.262*** −0.363*** (0.054) (0.056) Variable Fixed effects Geoadditive model Intercept −1.290*** (0.444) Ln(Mktpot) 0.096** 0.144*** (0.048) (0.047) Ln(Literacy) −0.262*** −0.363*** (0.054) (0.056) Notes: Estimation results of the parametric-fixed effect model and of the semiparametric geoadditive model. Coefficients and standard errors (in parenthesis). Number of observations: 276 (69 provinces by four points in time). Table A1 Whole manufacturing Variable Fixed effects Geoadditive model Intercept −1.290*** (0.444) Ln(Mktpot) 0.096** 0.144*** (0.048) (0.047) Ln(Literacy) −0.262*** −0.363*** (0.054) (0.056) Variable Fixed effects Geoadditive model Intercept −1.290*** (0.444) Ln(Mktpot) 0.096** 0.144*** (0.048) (0.047) Ln(Literacy) −0.262*** −0.363*** (0.054) (0.056) Notes: Estimation results of the parametric-fixed effect model and of the semiparametric geoadditive model. Coefficients and standard errors (in parenthesis). Number of observations: 276 (69 provinces by four points in time). A.4. Specialization, concentration and spatial dependence A.4.1. Krugman specialization index We address the question of how specialized were Italian provinces by using Krugman’s specialization index Ki that, for a given point in time (t), is defined as: Ki=∑k|(sik−si−k)|0≤Ki≤2, where sik=(vik)/(∑kvik), and si−k is the share of industry k in the national total net of province i. Thus, the Krugman index provides a measure of the difference in the specialization of a given province when compared with the remaining provinces of the country. It takes the value of 0 if there is no difference, and the value of 2 if a province has no industries in common with the rest of Italy. Figure A6 illustrates the territorial distribution of Ki in 1871 and 1911. Figure A6 View largeDownload slide Specialization: Krugman index. Figure A6 View largeDownload slide Specialization: Krugman index. A.4.2. Theil index of concentration The relative Theil index of concentration (Ck) is based on the normalized location quotient, that is LQik normalized by the ratio between vi (manufacturing value added in province i) over V (manufacturing value added in Italy): Ck=∑iviVln⁡(LQik). The higher the value of Ck, the higher the concentration of industry k. The relative Theil index Ck provides useful information about the extent to which industries are concentrated in a limited number of areas, but it does not take into consideration whether those areas are close together or far apart. In other words, it does not take into account the spatial structure of the data. Every region is treated as an island, and its position in space relative to other regions is not taken into account. Thus, the relative Theil index, Ck(t), is an a-spatial measure of concentration: the same degree of concentration can be compatible with very different localization schemes. For example, two industries may appear equally geographically concentrated, while one is located in two neighboring regions, and the other splits between the northern and the southern part of the country. As pointed out by Arbia et al. (2013), a more accurate analysis of the spatial distribution of economic activities requires the combination of traditional measures of geographical concentration and methodologies that account for spatial dependence, in that they provide different and complementary information about the concentration of the various sectors. A.4.3. Moran’s I index of spatial dependence Spatial autocorrelation is present when the values of one variable observed at nearby locations are more similar than those observed in locations that are far apart. More precisely, positive spatial autocorrelation occurs when high or low values of a variable tend to cluster together in space and negative spatial autocorrelation when high values are surrounded by low values and vice versa. Among the spatial dependence measures the most widely used is the Moran’s I index based, as is well known, on a comparison of LQik at any location with the value of the same variable at surrounding locations. The most widely used is the Moran’s I index: I=(N∑i∑jwij)(∑i∑jwij(LQi−LQ¯)(LQj−LQ¯)∑i(LQi−LQ¯)2), (4) where N is the total number of provinces, LQi and LQj are the observed values of the location quotient for the locations i and j (with mean LQ¯), and the first term is a scaling constant. This statistic compares the value of a continuous variable at any location with the value of the same variable at surrounding locations. The spatial structure of the data is formally expressed in a spatial weight matrix W with generic elements wij (with i≠j). In this paper, we employ a row-standardized spatial weights matrix (W), whose elements wij on the main diagonal are set to zero whereas wij = 1 if dij<d¯ and wij = 0 if dij>d¯, with dij the great circle distance between the centroids of region i and region j and d¯ a cut-off distance (equal to 112 km, corresponding to the minimum distance which allows all provinces to have at least one neighbor). Table A2 reports the calculated value of the Theil and Moran indices for the years 1871, 1881, 1901 and 1911 with data disaggregated by sector. Table A2 Industry concentration: Theil index and Moran’s I (1) (2) (3) (4) (5) (6) (7) (8) Theil Moran 1871 1881 1901 1911 1871 1881 1901 1911 2.1 Foodstuffs 0.02 0.02 0.04 0.06 0.15 0.31 0.25 0.29 (−0.02) (0.00) (0.00) (0.00) 2.2 Tobacco 1.23 1.05 1.04 0.81 −0.04 0.00 0.02 0.01 (−0.63) (−0.41) (−0.32) (−0.38) 2.3 Textiles 0.26 0.30 0.45 0.45 0.33 0.38 0.47 0.49 (0.00) (0.00) (0.00) (0.00)     2.3.1 Cotton 0.90 0.91 0.70 0.63 0.04 0.08 0.39 0.44 (0.23) (0.11) (0.00) (0.00)     2.3.2 Wool 1.09 1.21 1.40 1.30 0.03 0.02 0.00 0.00 (0.27) (0.27) (0.42) (0.41)     2.3.3 Silk 1.44 1.45 1.14 1.04 0.04 0.04 0.13 0.26 (0.04) (0.08) (0.00) (0.00)     2.3.4 Other natural fibers 0.27 0.37 0.23 0.40 0.19 0.45 0.07 −0.06 (0.00) (0.00) (0.14) (0.73) 2.4 Clothing 0.11 0.13 0.10 0.11 0.29 0.30 0.40 0.35 (0.00) (0.00) (0.00) (0.00) 2.5 Wood 0.02 0.02 0.03 0.05 0.11 0.10 0.22 0.37 (−0.05) (−0.06) (0.00) (0.00) 2.6 Leather 0.05 0.06 0.11 0.16 0.60 0.62 0.69 0.73 (0.00) (0.00) (0.00) (0.00) 2.7 Metalmaking 0.38 0.57 0.86 0.74 0.28 0.26 0.17 0.17 (0.00) (0.00) (0.00) (0.00) 2.8 Engineering 0.05 0.03 0.06 0.07 0.17 0.19 0.12 0.02 (−0.01) (0.00) (−0.03) (−0.34)     2.8.1 Shipbuildings 1.95 1.56 1.67 1.79 0.00 −0.01 −0.02 −0.06 (0.34) (0.49) (0.55) (0.78)     2.8.2 Machinery 0.16 0.09 0.22 0.19 0.02 0.35 0.06 0.08 (0.32) (0.00) (0.16) (0.10)     2.8.3 Blacksmith 0.04 0.03 0.04 0.02 0.35 0.37 0.43 0.11 (0.00) (0.00) (0.00) (0.05) 2.9 Non-metallic mineral products 0.17 0.17 0.19 0.09 0.20 0.08 0.12 0.11 (0.00) (−0.03) (0.00) (−0.02) 2.10 Chemicals and rubber 0.18 0.16 0.23 0.17 −0.04 0.15 0.00 0.05 (−0.61) (−0.02) (−0.40) (−0.19) 2.11 Paper and printing 0.21 0.21 0.17 0.16 0.11 0.07 0.06 0.07 (−0.05) (−0.12) (−0.17) (−0.13) 2.12 Sundry 0.33 0.89 0.62 0.46 0.05 0.00 0.01 0.15 (−0.19) (−0.37) (−0.36) (−0.01) 2. Total manufacturing 0.03 0.04 0.07 0.08 0.05 0.00 0.01 0.15 (−0.47) (−0.41) (−0.09) (−0.02) (1) (2) (3) (4) (5) (6) (7) (8) Theil Moran 1871 1881 1901 1911 1871 1881 1901 1911 2.1 Foodstuffs 0.02 0.02 0.04 0.06 0.15 0.31 0.25 0.29 (−0.02) (0.00) (0.00) (0.00) 2.2 Tobacco 1.23 1.05 1.04 0.81 −0.04 0.00 0.02 0.01 (−0.63) (−0.41) (−0.32) (−0.38) 2.3 Textiles 0.26 0.30 0.45 0.45 0.33 0.38 0.47 0.49 (0.00) (0.00) (0.00) (0.00)     2.3.1 Cotton 0.90 0.91 0.70 0.63 0.04 0.08 0.39 0.44 (0.23) (0.11) (0.00) (0.00)     2.3.2 Wool 1.09 1.21 1.40 1.30 0.03 0.02 0.00 0.00 (0.27) (0.27) (0.42) (0.41)     2.3.3 Silk 1.44 1.45 1.14 1.04 0.04 0.04 0.13 0.26 (0.04) (0.08) (0.00) (0.00)     2.3.4 Other natural fibers 0.27 0.37 0.23 0.40 0.19 0.45 0.07 −0.06 (0.00) (0.00) (0.14) (0.73) 2.4 Clothing 0.11 0.13 0.10 0.11 0.29 0.30 0.40 0.35 (0.00) (0.00) (0.00) (0.00) 2.5 Wood 0.02 0.02 0.03 0.05 0.11 0.10 0.22 0.37 (−0.05) (−0.06) (0.00) (0.00) 2.6 Leather 0.05 0.06 0.11 0.16 0.60 0.62 0.69 0.73 (0.00) (0.00) (0.00) (0.00) 2.7 Metalmaking 0.38 0.57 0.86 0.74 0.28 0.26 0.17 0.17 (0.00) (0.00) (0.00) (0.00) 2.8 Engineering 0.05 0.03 0.06 0.07 0.17 0.19 0.12 0.02 (−0.01) (0.00) (−0.03) (−0.34)     2.8.1 Shipbuildings 1.95 1.56 1.67 1.79 0.00 −0.01 −0.02 −0.06 (0.34) (0.49) (0.55) (0.78)     2.8.2 Machinery 0.16 0.09 0.22 0.19 0.02 0.35 0.06 0.08 (0.32) (0.00) (0.16) (0.10)     2.8.3 Blacksmith 0.04 0.03 0.04 0.02 0.35 0.37 0.43 0.11 (0.00) (0.00) (0.00) (0.05) 2.9 Non-metallic mineral products 0.17 0.17 0.19 0.09 0.20 0.08 0.12 0.11 (0.00) (−0.03) (0.00) (−0.02) 2.10 Chemicals and rubber 0.18 0.16 0.23 0.17 −0.04 0.15 0.00 0.05 (−0.61) (−0.02) (−0.40) (−0.19) 2.11 Paper and printing 0.21 0.21 0.17 0.16 0.11 0.07 0.06 0.07 (−0.05) (−0.12) (−0.17) (−0.13) 2.12 Sundry 0.33 0.89 0.62 0.46 0.05 0.00 0.01 0.15 (−0.19) (−0.37) (−0.36) (−0.01) 2. Total manufacturing 0.03 0.04 0.07 0.08 0.05 0.00 0.01 0.15 (−0.47) (−0.41) (−0.09) (−0.02) Note: p-values in parenthesis. Table A2 Industry concentration: Theil index and Moran’s I (1) (2) (3) (4) (5) (6) (7) (8) Theil Moran 1871 1881 1901 1911 1871 1881 1901 1911 2.1 Foodstuffs 0.02 0.02 0.04 0.06 0.15 0.31 0.25 0.29 (−0.02) (0.00) (0.00) (0.00) 2.2 Tobacco 1.23 1.05 1.04 0.81 −0.04 0.00 0.02 0.01 (−0.63) (−0.41) (−0.32) (−0.38) 2.3 Textiles 0.26 0.30 0.45 0.45 0.33 0.38 0.47 0.49 (0.00) (0.00) (0.00) (0.00)     2.3.1 Cotton 0.90 0.91 0.70 0.63 0.04 0.08 0.39 0.44 (0.23) (0.11) (0.00) (0.00)     2.3.2 Wool 1.09 1.21 1.40 1.30 0.03 0.02 0.00 0.00 (0.27) (0.27) (0.42) (0.41)     2.3.3 Silk 1.44 1.45 1.14 1.04 0.04 0.04 0.13 0.26 (0.04) (0.08) (0.00) (0.00)     2.3.4 Other natural fibers 0.27 0.37 0.23 0.40 0.19 0.45 0.07 −0.06 (0.00) (0.00) (0.14) (0.73) 2.4 Clothing 0.11 0.13 0.10 0.11 0.29 0.30 0.40 0.35 (0.00) (0.00) (0.00) (0.00) 2.5 Wood 0.02 0.02 0.03 0.05 0.11 0.10 0.22 0.37 (−0.05) (−0.06) (0.00) (0.00) 2.6 Leather 0.05 0.06 0.11 0.16 0.60 0.62 0.69 0.73 (0.00) (0.00) (0.00) (0.00) 2.7 Metalmaking 0.38 0.57 0.86 0.74 0.28 0.26 0.17 0.17 (0.00) (0.00) (0.00) (0.00) 2.8 Engineering 0.05 0.03 0.06 0.07 0.17 0.19 0.12 0.02 (−0.01) (0.00) (−0.03) (−0.34)     2.8.1 Shipbuildings 1.95 1.56 1.67 1.79 0.00 −0.01 −0.02 −0.06 (0.34) (0.49) (0.55) (0.78)     2.8.2 Machinery 0.16 0.09 0.22 0.19 0.02 0.35 0.06 0.08 (0.32) (0.00) (0.16) (0.10)     2.8.3 Blacksmith 0.04 0.03 0.04 0.02 0.35 0.37 0.43 0.11 (0.00) (0.00) (0.00) (0.05) 2.9 Non-metallic mineral products 0.17 0.17 0.19 0.09 0.20 0.08 0.12 0.11 (0.00) (−0.03) (0.00) (−0.02) 2.10 Chemicals and rubber 0.18 0.16 0.23 0.17 −0.04 0.15 0.00 0.05 (−0.61) (−0.02) (−0.40) (−0.19) 2.11 Paper and printing 0.21 0.21 0.17 0.16 0.11 0.07 0.06 0.07 (−0.05) (−0.12) (−0.17) (−0.13) 2.12 Sundry 0.33 0.89 0.62 0.46 0.05 0.00 0.01 0.15 (−0.19) (−0.37) (−0.36) (−0.01) 2. Total manufacturing 0.03 0.04 0.07 0.08 0.05 0.00 0.01 0.15 (−0.47) (−0.41) (−0.09) (−0.02) (1) (2) (3) (4) (5) (6) (7) (8) Theil Moran 1871 1881 1901 1911 1871 1881 1901 1911 2.1 Foodstuffs 0.02 0.02 0.04 0.06 0.15 0.31 0.25 0.29 (−0.02) (0.00) (0.00) (0.00) 2.2 Tobacco 1.23 1.05 1.04 0.81 −0.04 0.00 0.02 0.01 (−0.63) (−0.41) (−0.32) (−0.38) 2.3 Textiles 0.26 0.30 0.45 0.45 0.33 0.38 0.47 0.49 (0.00) (0.00) (0.00) (0.00)     2.3.1 Cotton 0.90 0.91 0.70 0.63 0.04 0.08 0.39 0.44 (0.23) (0.11) (0.00) (0.00)     2.3.2 Wool 1.09 1.21 1.40 1.30 0.03 0.02 0.00 0.00 (0.27) (0.27) (0.42) (0.41)     2.3.3 Silk 1.44 1.45 1.14 1.04 0.04 0.04 0.13 0.26 (0.04) (0.08) (0.00) (0.00)     2.3.4 Other natural fibers 0.27 0.37 0.23 0.40 0.19 0.45 0.07 −0.06 (0.00) (0.00) (0.14) (0.73) 2.4 Clothing 0.11 0.13 0.10 0.11 0.29 0.30 0.40 0.35 (0.00) (0.00) (0.00) (0.00) 2.5 Wood 0.02 0.02 0.03 0.05 0.11 0.10 0.22 0.37 (−0.05) (−0.06) (0.00) (0.00) 2.6 Leather 0.05 0.06 0.11 0.16 0.60 0.62 0.69 0.73 (0.00) (0.00) (0.00) (0.00) 2.7 Metalmaking 0.38 0.57 0.86 0.74 0.28 0.26 0.17 0.17 (0.00) (0.00) (0.00) (0.00) 2.8 Engineering 0.05 0.03 0.06 0.07 0.17 0.19 0.12 0.02 (−0.01) (0.00) (−0.03) (−0.34)     2.8.1 Shipbuildings 1.95 1.56 1.67 1.79 0.00 −0.01 −0.02 −0.06 (0.34) (0.49) (0.55) (0.78)     2.8.2 Machinery 0.16 0.09 0.22 0.19 0.02 0.35 0.06 0.08 (0.32) (0.00) (0.16) (0.10)     2.8.3 Blacksmith 0.04 0.03 0.04 0.02 0.35 0.37 0.43 0.11 (0.00) (0.00) (0.00) (0.05) 2.9 Non-metallic mineral products 0.17 0.17 0.19 0.09 0.20 0.08 0.12 0.11 (0.00) (−0.03) (0.00) (−0.02) 2.10 Chemicals and rubber 0.18 0.16 0.23 0.17 −0.04 0.15 0.00 0.05 (−0.61) (−0.02) (−0.40) (−0.19) 2.11 Paper and printing 0.21 0.21 0.17 0.16 0.11 0.07 0.06 0.07 (−0.05) (−0.12) (−0.17) (−0.13) 2.12 Sundry 0.33 0.89 0.62 0.46 0.05 0.00 0.01 0.15 (−0.19) (−0.37) (−0.36) (−0.01) 2. Total manufacturing 0.03 0.04 0.07 0.08 0.05 0.00 0.01 0.15 (−0.47) (−0.41) (−0.09) (−0.02) Note: p-values in parenthesis. A.5. The sectorial effects of Alpine regions as a function of river The following figure shows the estimated marginal effects of the dummy Alpine, conditional on the effect of market potential and literacy, as a function of the river variable. Alpine provinces show a comparative advantage in the location of industrial activity in the case of machinery, cotton, silk, blacksmith and non-metallic mineral products. Figure A7 View largeDownload slide Estimated coefficient of Alpine by ln(River) with simulated 95% confidence intervals. Figure A7 View largeDownload slide Estimated coefficient of Alpine by ln(River) with simulated 95% confidence intervals. © The Author (2017). Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of Economic Geography Oxford University Press

The location of the Italian manufacturing industry, 1871–1911: a sectoral analysis

Basile, Roberto; Ciccarelli, Carlo
Journal of Economic Geography , Volume Advance Article (3) – Oct 9, 2017
Download PDF
35 pages

Loading next page...
 
/lp/ou_press/the-location-of-the-italian-manufacturing-industry-1871-1911-a-lRpuuCGNaN
Publisher
Oxford University Press
Copyright
© The Author (2017). Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com
ISSN
1468-2702
eISSN
1468-2710
DOI
10.1093/jeg/lbx033
Publisher site
See Article on Publisher Site

Abstract

Abstract This study focuses on industrial location in Italy during the period 1871–1911, when manufacturing moved from artisanal to factory-based production processes. There is general agreement in the historical and economic literature that factor endowment and domestic market potential represented the main drivers of industrial location. We test the relative importance of the above drivers of location for the various manufacturing sectors using data at the provincial level. Estimation results reveal that the location of capital intensive sectors (such as chemicals, cotton, metalmaking and paper) was driven by domestic market potential and literacy. Once market potential and literacy are accounted for, the evidence on the effect of water endowment on industrial location is mixed, depending on the manufacturing sector considered. 1. Introduction Economic theory suggests that factor endowment and market access are the key determinants of industrial location. On the one hand, the neoclassical trade theory predicts that regional differences in factor endowments (such as mineral deposits, water supply and labor skills) contribute to determine regional comparative advantages and, therefore, regional specialization. On the other hand, the New Economic Geography (NEG) literature (Krugman, 1991; Fujita et al., 1999) suggests that an uneven spatial distribution of market access encourages firms to concentrate in regions with higher market potential, to benefit from increasing returns and to export goods and services to other regions. The relative effect of factor endowment and market access on industrial location can be hardly quantified using empirical data referring to modern economies, as the two forces tend to coexist and interact in a very complex way along with the effect of (endogenous) policy interventions. Midelfart-Knarvik and Overman (2002) show for instance that the European industrial policy strongly influenced the industrial location patterns across the EU regions. As long as industrial policies are endogenously driven by the actual spatial distribution of economic activities, it can be quite difficult to quantify the genuine (net of industrial policies) effect of comparative advantages and/or market potential on industrial location. However, the use of historical data covering the years following the political unification of a country (i.e., when the process of domestic market integration was taking its first steps, and no systematic regional industrial policy existed) may provide an opportunity to better contrast the different explanations for the spatial concentration of industry and, in particular, to appreciate the role of the home-market effect. Industrialization processes, the fall in transport costs and the integration of domestic markets may indeed generate the agglomeration forces that change the distribution of economic activities across space and reinforce spatial disparities over time. As a matter of fact, the economic history literature has increasingly drawn on NEG models to analyze national historical experiences and, due to its rising importance, the manufacturing sector has received most of the attention. Examples of studies in this direction are Wolf (2007) for Poland, Klein and Crafts (2012) for the USA, Rosés (2003), Tirado et al. (2002), and Martinez-Galarraga (2012) for Spain and Crafts and Mulatu (2005) for Britain. As far as Italy is concerned, A’Hearn and Venables (2013) explore the interactions between external trade and regional disparities since the unification of the country (in 1861). The authors argue that the economic superiority of Northern regions over the rest of the country was initially based on natural advantages (in particular the endowment of water), while from the late 1880s onwards domestic market access became a key determinant of industrial location, inducing dynamic industrial sectors to locate in regions with a large domestic market—that is, in the North. From 1945 onwards, however, with the gradual process of European integration, foreign market access became the decisive factor; and the North, once again, had the advantage of proximity to these markets. While we broadly agree with the periodization proposed by A’Hearn and Venables (2013), we believe that it requires further qualification. Specifically, in line with Rosés (2003) for Spain and Klein and Crafts (2012) for the USA, we believe that the relative importance of factor endowment and market potential for industrial location varies according to the technology prevailing in the various sectors. In particular, market potential is expected to be more important in industries characterized by increasing returns to scale (typically, high and medium capital intensive industries), while factor endowment should be more relevant in the remaining manufacturing sectors. On the basis of these considerations, we analyze the spatial location patterns of the various branches of the Italian manufacturing industry during the period 1871–1911. Specifically, we assess the relative importance of factor endowment (water abundance and labor skills) and domestic market potential for industrial location behavior in the early phases of Italian industrialization, distinguishing among the various manufacturing sectors. Our analysis is based on sectoral value added data at 1911 prices at the provincial level, recently produced by Ciccarelli and Fenoaltea (2013, 2014). Our results clearly show that as transportation costs decreased and institutional barriers to domestic trade were eliminated, Italian provinces became more and more specialized and manufacturing activity became increasingly concentrated in a few provinces, mostly belonging to the North-West. The estimation results corroborate the hypothesis that both comparative advantages and domestic market potential have been responsible for this process of spatial concentration. However, the econometric analysis also reveals a heterogeneous reaction of the location of various branches of manufacturing to domestic market potential and factor endowment. In particular, the location of medium and high capital intensive sectors (including chemicals, cotton, metalmaking, machinery and paper) was mainly driven by the domestic market potential and literacy. Water endowment represented an important driver of industrial location for sectors of heterogeneous nature (such as chemicals, silk and leather). The rest of the paper is organized as follows. Section 2 illustrates the spatial distribution of manufacturing industry over the sample period. Section 3 describes the hypotheses on the key drivers of industrial location. Section 4 reports the estimation results. Section 5 concludes. 2. The spatial diffusion of manufacturing activity in Italy: 1871–1911 2.1. Setting the scene Figure 1 illustrates the division of the Italian territory that roughly prevailed during the 1815–1860 period. The seven pre-unitarian states (Kingdom of Sardinia, Kingdom of Lombardy-Venetia, Duchy of Parma and Piacenza, Duchy of Modena and Reggio, Grand Duchy of Tuscany, Papal States and Kingdom of the Two Sicilies) were characterized by extremely different institutions and economic policies. The coin, monetary regimes and trade policies were different. Primary schooling was mandatory only in certain pre-unitarian states, mainly those in the North. Figure 1 View largeDownload slide Italian provinces at 1911 borders grouped into pre-unitarian states (1815–1860 ca.). Figure 1 View largeDownload slide Italian provinces at 1911 borders grouped into pre-unitarian states (1815–1860 ca.). Italy was unified in 1861, although Venetia and Latium were annexed to the country only in 1866 and 1870, respectively. Between 1861 and 1870, the national capital was moved from Turin to Florence and, finally, to Rome. Soon after the political unification, policy makers realized that there was an urgent need of statistical information. Decennial population censuses were established, and dozens of annual reports to the Italian Parliament and other official publications concerning the main economic sectors (public budget and taxation, international trade, railroads, public school system) were regularly produced. The new official historical statistics divided the Italian territory into 16 regions (compartimenti, roughly NUTS-2 units) and 69 provinces (province, roughly NUTS-3 units). The borders of these administrative units (shown in Figure 1 and in the Appendix) did not change between 1871 and 1911. 2.2. The structure of economic activity Between 1871 and 1911, the composition of Italy’s GDP changed considerably, with industry increasing (from about 15% to 25%) at the expense of agriculture. At the same time, a considerable sectoral reallocation within manufacturing occurred. Table 1 reports the sectoral distribution of manufacturing value added at benchmark years (1871, 1881, 1901, and 1911), when population censuses were taken. With few exceptions, sectors tied to the production of consumption goods (roughly those from 2.1 foodstuffs to 2.6 leather) show a constant reduction in their shares. Sectors tied to the production of durable goods (roughly those from 2.7 metalmaking to 2.11 paper) follow an opposite long-term trend, with a rapid acceleration in the 1901–1911 decade. The last column of Table 1 summarizes the 1871–1911 trends, with numbers below 1 indicating a reduction of the sectoral share, and vice versa. Table 1 Manufacturing sectors: value added sharesa (percentages) Sectors 1871 1881 1901 1911 1911/1871 2.1 Foodstuffs 33.5 30.4 25.4 21.5 0.6 2.2 Tobacco 1.6 1.3 0.9 0.7 0.4 2.3 Textiles 10.3 10.3 12.8 11.1 1.1     2.3.1 Cotton 1.3 1.9 5.2 4.8 3.7     2.3.2 Wool 1.8 2.0 2.4 2.3 1.3     2.3.3 Silk 3.9 3.8 4.3 3.3 0.8     2.3.4 Other natural fibers 3.2 2.5 0.9 0.7 0.2 2.4 Clothing 6.9 7.4 6.8 6.3 0.9 2.5 Wood 10.0 9.3 9.7 10.0 1.0 2.6 Leather 10.5 11.5 11.4 7.8 0.7 2.7 Metalmaking 0.6 1.0 1.7 3.1 5.2 2.8 Engineering 17.5 17.8 18.7 21.5 1.2     2.8.1 Shipbuilding 2.0 1.4 3.0 1.9 1.0     2.8.2 Machinery 2.9 4.3 7.4 10.5 3.6     2.8.3 Blacksmith 12.5 12.2 9.1 9.0 0.7 2.9 Non-metallic mineral products 3.6 4.2 4.2 6.6 1.8 2.10 Chemicals and rubber 2.1 2.4 3.0 4.3 2.0 2.11 Paper 2.7 3.5 4.8 6.3 2.3 2.12 Sundry 0.7 0.7 0.6 0.7 1.0 2. Total manufacturing 100.0 100.0 100.0 100.0 1.0 Sectors 1871 1881 1901 1911 1911/1871 2.1 Foodstuffs 33.5 30.4 25.4 21.5 0.6 2.2 Tobacco 1.6 1.3 0.9 0.7 0.4 2.3 Textiles 10.3 10.3 12.8 11.1 1.1     2.3.1 Cotton 1.3 1.9 5.2 4.8 3.7     2.3.2 Wool 1.8 2.0 2.4 2.3 1.3     2.3.3 Silk 3.9 3.8 4.3 3.3 0.8     2.3.4 Other natural fibers 3.2 2.5 0.9 0.7 0.2 2.4 Clothing 6.9 7.4 6.8 6.3 0.9 2.5 Wood 10.0 9.3 9.7 10.0 1.0 2.6 Leather 10.5 11.5 11.4 7.8 0.7 2.7 Metalmaking 0.6 1.0 1.7 3.1 5.2 2.8 Engineering 17.5 17.8 18.7 21.5 1.2     2.8.1 Shipbuilding 2.0 1.4 3.0 1.9 1.0     2.8.2 Machinery 2.9 4.3 7.4 10.5 3.6     2.8.3 Blacksmith 12.5 12.2 9.1 9.0 0.7 2.9 Non-metallic mineral products 3.6 4.2 4.2 6.6 1.8 2.10 Chemicals and rubber 2.1 2.4 3.0 4.3 2.0 2.11 Paper 2.7 3.5 4.8 6.3 2.3 2.12 Sundry 0.7 0.7 0.6 0.7 1.0 2. Total manufacturing 100.0 100.0 100.0 100.0 1.0 aThe table includes 12 manufacturing sectors (numbered from 2.1 to 2.12, as it is customary in the national account, where usually 1 refers to the extractive sector, 2 to manufacturing, 3 to constructions and 4 to the utilities). Textiles (2.3) and engineering (2.8) are further disaggregated, as detailed in the Appendix. Numbers need not to add due to rounding. Table 1 Manufacturing sectors: value added sharesa (percentages) Sectors 1871 1881 1901 1911 1911/1871 2.1 Foodstuffs 33.5 30.4 25.4 21.5 0.6 2.2 Tobacco 1.6 1.3 0.9 0.7 0.4 2.3 Textiles 10.3 10.3 12.8 11.1 1.1     2.3.1 Cotton 1.3 1.9 5.2 4.8 3.7     2.3.2 Wool 1.8 2.0 2.4 2.3 1.3     2.3.3 Silk 3.9 3.8 4.3 3.3 0.8     2.3.4 Other natural fibers 3.2 2.5 0.9 0.7 0.2 2.4 Clothing 6.9 7.4 6.8 6.3 0.9 2.5 Wood 10.0 9.3 9.7 10.0 1.0 2.6 Leather 10.5 11.5 11.4 7.8 0.7 2.7 Metalmaking 0.6 1.0 1.7 3.1 5.2 2.8 Engineering 17.5 17.8 18.7 21.5 1.2     2.8.1 Shipbuilding 2.0 1.4 3.0 1.9 1.0     2.8.2 Machinery 2.9 4.3 7.4 10.5 3.6     2.8.3 Blacksmith 12.5 12.2 9.1 9.0 0.7 2.9 Non-metallic mineral products 3.6 4.2 4.2 6.6 1.8 2.10 Chemicals and rubber 2.1 2.4 3.0 4.3 2.0 2.11 Paper 2.7 3.5 4.8 6.3 2.3 2.12 Sundry 0.7 0.7 0.6 0.7 1.0 2. Total manufacturing 100.0 100.0 100.0 100.0 1.0 Sectors 1871 1881 1901 1911 1911/1871 2.1 Foodstuffs 33.5 30.4 25.4 21.5 0.6 2.2 Tobacco 1.6 1.3 0.9 0.7 0.4 2.3 Textiles 10.3 10.3 12.8 11.1 1.1     2.3.1 Cotton 1.3 1.9 5.2 4.8 3.7     2.3.2 Wool 1.8 2.0 2.4 2.3 1.3     2.3.3 Silk 3.9 3.8 4.3 3.3 0.8     2.3.4 Other natural fibers 3.2 2.5 0.9 0.7 0.2 2.4 Clothing 6.9 7.4 6.8 6.3 0.9 2.5 Wood 10.0 9.3 9.7 10.0 1.0 2.6 Leather 10.5 11.5 11.4 7.8 0.7 2.7 Metalmaking 0.6 1.0 1.7 3.1 5.2 2.8 Engineering 17.5 17.8 18.7 21.5 1.2     2.8.1 Shipbuilding 2.0 1.4 3.0 1.9 1.0     2.8.2 Machinery 2.9 4.3 7.4 10.5 3.6     2.8.3 Blacksmith 12.5 12.2 9.1 9.0 0.7 2.9 Non-metallic mineral products 3.6 4.2 4.2 6.6 1.8 2.10 Chemicals and rubber 2.1 2.4 3.0 4.3 2.0 2.11 Paper 2.7 3.5 4.8 6.3 2.3 2.12 Sundry 0.7 0.7 0.6 0.7 1.0 2. Total manufacturing 100.0 100.0 100.0 100.0 1.0 aThe table includes 12 manufacturing sectors (numbered from 2.1 to 2.12, as it is customary in the national account, where usually 1 refers to the extractive sector, 2 to manufacturing, 3 to constructions and 4 to the utilities). Textiles (2.3) and engineering (2.8) are further disaggregated, as detailed in the Appendix. Numbers need not to add due to rounding. In 1911, foodstuffs, textiles and engineering alone represented more than 50% of total value added in manufacturing. Effectively, Ciccarelli and Proietti (2013) show that these three sectors explain much of the variability of sectoral specialization at the provincial level, and thus act as a sort of ‘sufficient statistic’ for the whole manufacturing industry during 1871–1911. In addition, Fenoaltea (2016) documents how vast and heterogeneous was the engineering sectors and Rosés (2003) bases his results for Spain on the separated components of the textile industry. For these reasons, we disaggregated the data on value added for the textile and engineering sectors into sub-sectors (this was not possible for foodstuffs due to the lack of historical data). Table 1 clearly shows the importance of this disaggregation. Within the textile sector, only cotton, more suitable to mechanization than other fibers, increased substantially its value added share over time. Within engineering, the size of the machinery component increased substantially. At the same time, however, traditional blacksmith activities, despite a declining trend, accounted for about half of value added of the engineering sector even at the end of our sample period. After all, 19th-century Italy was very much an agricultural country and the maintenance of agricultural tools (such as spades, hoes and ploughs) represented a substantial part of blacksmiths’ traditional activity. 2.3. Regional specialization and geographical concentration of manufacturing activity To analyze the dynamics of regional specialization, industrial concentration and the spatial distribution of industrial activities during 1871–1911, we use disaggregated data on manufacturing value added, vik(t), at 1911 prices in province i ( i=1,…,69), sector k ( k=1,…,17)and time t ( t=1871,1881,1901,1911) and compute, for each time t the location quotient LQik=(vik/∑kvik)/(∑kvik/∑i∑kvik) (see the Appendix for details on data and indices). In 1871, the manufacturing industry mainly clustered in a few Northern provinces (Figure 2). To some extent, it was however also present in the South. Naples and Palermo, the provinces with the ancient capitals of the Kingdom of the Two Sicilies, registered in particular noticeable LQ values (respectively, 1.41 and 1.32) in line with Turin (1.35) and Genoa (1.20) that, together with Milan (1.57), formed the vertices of the so-called ‘industrial triangle’. In 1911, the North-West reinforced its dominant position, while most of the Southern provinces worsened their relative position. Figure 2 View largeDownload slide Manufacturing: choropleth maps of LQ values. Figure 2 View largeDownload slide Manufacturing: choropleth maps of LQ values. The increase in spatial inequalities in Italian manufacturing activity is also revealed by the values of the Theil index of concentration (passing from 0.03 in 1871 to 0.08 in 1911) and of the Moran’s I index of spatial autocorrelation (passing from a statically non-significant negative value of −0.04 in 1871 to a significant value of 0.05 in 1911). Moreover, the average degree of industrial specialization of Italian provinces (measured by the Krugman index) increased monotonically over the sample period (passing from 0.26 to 0.38). The Krugman index increased particularly for selected provinces of the North-West (including Turin, Milan, Genoa, Novara, Como, Cremona and Bergamo) and Tuscany (Massa Carrara, Lucca, Pisa and Leghorn) (see the Appendix). These findings are in line with the predictions of Core-Periphery NEG models (Krugman, 1991), according to which, when interregional trade costs decrease,1 industrial activities characterized by increasing returns to scale tend to concentrate in the core regions (i.e., those with higher demand; home-market effect), while the remaining (peripheral) regions suffer de-industrialization. Increased competition, in the labor market for instance, may, however, create an inverted U-shape pattern: as trade costs continue to fall, a second stage of adjustment could occur when market forces operate to undermine the core–periphery pattern and reduce regional inequalities (Krugman and Venables, 1995; Puga, 1999). However, this second stage of adjustment refers to the most recent history of globalization, so we may conclude that the evolution of the spatial distribution of the overall manufacturing activity during the period 1871–1911 is a good fit for the predictions of the core–periphery model. The spatial diffusion of the overall manufacturing activity is nevertheless the net result of heterogeneous dynamics across the 17 manufacturing branches. To analyze sectoral developments, we combine the a-spatial Theil index and the Moran’s I index in a scatterplot for 1871 and 1911, excluding tobacco and sundry (Figure 3). The vertical and horizontal dashed lines denote median values and identify four patterns in the distribution of economic activities: (a) HL (high concentration and low spatial dependence), (b) HH (high concentration and high spatial dependence), (c) LH (low concentration and high spatial dependence) and (d) LL (low concentration and low spatial dependence). Figure 3 View largeDownload slide Spatial concentration: scatterplot between the a-spatial Theil index of concentration (y-axis), and the Moran index of spatial autocorrelation (x-axis). Figure 3 View largeDownload slide Spatial concentration: scatterplot between the a-spatial Theil index of concentration (y-axis), and the Moran index of spatial autocorrelation (x-axis). Shipbuilding and leather appear the most extreme cases. Shipbuilding belongs both in 1871 and 1911 to spatial pattern a (high Theil, low Moran). This means that this industry concentrated its activity in a small number of (coastal) areas that were not close to each other. In this kind of sector, indeed, economies of scale are reached by increasing the plant size and concentrating the production in a small number of locations. Leather, both in 1871 and 1911, belongs to spatial pattern c (low Theil and high Moran). Other sectors kept their position over time. Capital intensive metalmaking belongs in 1871 and 1911 to spatial pattern b (high concentration and high spatial dependence). Traditional economic activities tied to the agricultural sector such as clothing, foodstuffs and wood, but also blacksmiths and non-metallic mineral products belong, although with some difference, to spatial pattern c (as does leather). Within textiles, silk and cotton, always characterized by above median Theil index, moved from spatial pattern a to spatial pattern b, becoming strongly agglomerated in 1911. On the contrary the other natural fibers components moved from spatial pattern b to spatial pattern a. The reallocation of the textile industry is consistent with Fenoaltea (2011) for Italy. The above evidence is also in line with the international literature. Rosés (2003) and Wolf (2007) document, respectively, for 19th-century Spain and for Poland in 1925–1937 that metalmaking and textiles were increasingly concentrated. Within the engineering sector, shipbuilding, machinery and blacksmith differ tremendously in terms of concentration, but also on the spatial autocorrelation dimension. This confirms the importance of considering further disaggregated data for this industry. One further notes that fast growing sectors (such as machinery, paper and chemicals) present in 1911 about median values of Theil and Moran. Finally, one notices a substantial continuity in the ranking of sectors along the vertical dimension (with traditional sectors like foodstuffs, clothing, blacksmiths, but also wood being the most dispersed sectors). Indeed, the ranking of the Theil indices remained very stable over time (the rank correlation between the Theil indices in 1871 and in 1911 is 0.95 with a p-value of 0.000). Interestingly, the stability in the level of concentration across industries is a pattern also found in Crafts and Mulatu (2006) for 19th-century Britain. So far we have documented the significant specialization and concentration of manufacturing industry that occurred during the sample period. In the following section, we will explore the potential drivers of this spatial evolution. 3. Drivers of industrial location The literature analyzing the main determinants of regional industrial location in the late 19th and the early 20th centuries focuses on the role played by factor endowment and market potential. Crafts and Mulatu (2006) show that patterns of industrial location in Britain were rather persistent and only marginally affected by falling transport costs during 1871–1911. Factor endowment (coal abundance) had a stronger effect on overall industrial location than proximity to markets. However, the latter was an attraction for industries with large plant size, above all shipbuilding and textiles. In addition, educated workers were an important input for the chemical but not for the textile sector. Klein and Crafts (2012) consider industrial location in the USA during 1880–1920 and show that both agglomeration mechanisms related to market potential and natural advantages influenced industrial location, although the former were increasingly important for sectors where plant size was relatively large, mainly located in the manufacturing belt. Rosés (2003) analyzes 19th-century Spain and shows that, as transport costs decreased and barriers to domestic trade were eliminated, Spanish manufacturing industry became increasingly concentrated in a few regions. Both market potential and factor endowment contribute to explain industrial location. However, the home-market effect played a key role for modern industries producing heterogeneous goods and experiencing increasing returns to scale. Wolf (2007) shows that skilled labor and inter-industry linkages have an important explanatory power for the industrial location in the case of 1925–1937 Poland. Inspired by this literature, in the remaining part of this section we provide possible measures of factor endowment and market potential and posit some hypotheses on their role as main determinants of industrial location in the case of 19th-century Italy. 3.1. Factor endowment 3.1.1. Water supply The historical literature (e.g., Cafagna, 1989) stressed the central role of natural endowment (above all, water) for industrial location in 19th-century Italy. The point was recently reiterated by Fenoaltea (2011), providing perhaps the most careful and sharp account of Italian industrialization over the 1861–1913 period. According to Fenoaltea, ‘the roots of the success of the Northern regions seem […] environmental rather than historical’ (p. 231): thus, factor endowment, more than socioeconomic variables such as social capital or institutions. Modern (or factory-based) systems of production gradually replaced artisanal production and ‘gave a strong advantage to the locations with a year-round supply of water (for power and also in the specific case of textiles, for the repeated washing of the material); and in Italy such locations abound only on the northern edge of the Po valley, where the Alpine run-off offsets the lack of summer rain’. Moreover, when analyzing the location dynamics of the textile industry during 1871–1911, Fenoaltea (2011, 232) mentions among the natural advantages of northern regions ‘the water that flowed in the rivers, the water suspended in the air […]’ and the presence of ‘mountain glaciers’. Beyond representing a source of motive power and adequate climate for textile industry (as stressed by Italy’s economic historians), rivers are in principle important determinants of industrial location of most manufacturing activities. Water may indeed work as a production input for washing, cooling, boiling and so on for both modern industries (such as chemicals) and more traditional industrial activities. As an example of the latter, consider the case of leather. Each step of the manufacturing process to obtain leather from the skins of the animals (tanning, retanning and finishing) requires a considerable amount of fresh water (Sundar et al., 2001). These arguments, and the lack of a comprehensive supporting literature on the matter, suggest that it is virtually impossible to posit specific hypotheses on the (possibly heterogeneous) response of the location of the various manufacturing sectors to water abundance. In principle, any sector should have benefited from the presence of rivers. So, we will essentially let the data speak for themselves, and assess through the econometric analysis whether water supply significantly affected industrial location net of the effect of the other variables included in the model (i.e., market potential and literacy). To measure water endowment at provincial level, the present study considers two time-invariant variables. The first one is the number of rivers in a province weighted by their economic ‘relevance’. The historical source Annuario Statistico Italiano (1886, 22–27) provides an exhaustive list of rivers flowing through the 69 Italian provinces. These rivers are ranked, with a value ranging from 1 to 5, according to the weight assigned to them by the experts of the Italian Military Geographic Institute (IGM) in a publication that culminates a research project that started in the 1960s (IGM, 2007). The importance of each river is established on the basis of ‘the length of each river and its socio-economic relevance’ (IGM, 2007, preface). In the IGM ranking, a value of 1 means ‘high relevance’, while a value of 5 means ‘low relevance’. Thus, our first measure of water endowment in each province i is the weighted average of the number of rivers in the province using (1−IGMrank/6) as weights: RIVERi=∑r∈i(1−IGMrankr/6), where r refers to each river flowing through province i. The second variable is a dummy indicating whether the province belongs to the Alpine region. Italian rivers are essentially of two types: Alpine rivers and Apennine rivers. Alpine rivers typically descend from the Alps, flowing from the North into the upper bank of the Po river. The Alps work in a sense as a sponge that absorbs water in the autumn and winter, and releases water in the spring and summer, when the glaciers melt. Alpine rivers are therefore rich in water throughout the year, while the Apennine rivers are relatively dry during the summer season—limiting the regular development of factory-based manufacturing activities that are not organized on a seasonal basis. In addition, Alpine rivers also have a higher flow rate than the Apennine rivers. As we will document in our econometric analysis, the interaction between the continuous variable RIVERi and the Alpine region dummy variable proved to be of significant help for our understanding of industrial location of certain manufacturing sectors, and supports nicely the need for industrial use of a year-round supply of water considered by Fenoaltea (2011). The provincial distribution of these water-related variables is illustrated in Figure 4. Figure 4 View largeDownload slide Water endowment. (a) Rivers and (b) Alpine region. Panel (a) shows the geographical distribution of the RIVERi variable, measuring the (weighted) number of rivers flowing through province i; panel (b) shows the provinces belonging to the Alpine region, located on the left (northern) bank of the Po river. Figure 4 View largeDownload slide Water endowment. (a) Rivers and (b) Alpine region. Panel (a) shows the geographical distribution of the RIVERi variable, measuring the (weighted) number of rivers flowing through province i; panel (b) shows the provinces belonging to the Alpine region, located on the left (northern) bank of the Po river. 3.1.2. Literacy Data on literacy rates (share of individuals aged 6 years and above able to read and write) at the provincial level for the years 1871, 1881, 1901 and 1911 are those reported in the population censuses. Data on literacy rates for 1861 (used, as we will illustrate, as an instrument for literacy rates in 1871) are also from the population census. However, in 1861 Latium and Venetia were not part of Italy and census data are not available. To fill this gap, we proceed as follows. Estimates of literacy rates for the provinces of Venetia in 1861 are obtained by assuming a constant 1871–1861 ratio of literacy rates in the regions of Lombardy and Venetia (both part of the Habsburg Empire during 1815–1860 ca.). Similarly, the estimates of literacy rates for Latium in 1861 assume a constant 1871–1861 ratio of literacy rates of Latium on the one hand, and of the macro-area formed by Emilia, Umbria and the Marches on the other hand. The underlying assumption is that, between 1861 and 1871, literacy rates of the regions forming the Papal States (during 1815–1860 ca.) evolved similarly. Figure 5 illustrates the geographical distribution of literacy rates in 1871 and 1911. As a result of very different socioeconomic developments and pre-unitarian policies, the North-South regional divide is particularly marked (Ciccarelli and Weisdorf, 2016). In 1871, the north-western regions registered literacy rates of about 50%, central and north-eastern regions of about 30% and southern regions of about 15%. In 1911, after four decades of mandatory primary public schooling, literacy rates reached some 75% in the north-west, 55 in the north-east and in the center and about 35% in the south. Figure 5 View largeDownload slide Literacy rates. Share of population aged 6 years and above able to read and write. Figure 5 View largeDownload slide Literacy rates. Share of population aged 6 years and above able to read and write. 3.2. Market potential In order to analyze the importance of domestic market access as a key driver of the spatial distribution of economic activity, a sound measure of accessibility to demand is required. In line with Klein and Crafts (2012), we construct market potential estimates for each Italian province i between 1871 and 1911 using Harris (1954)’s formula, that is, as a weighted average of GDP (or total value added) of all provinces j2: MKTPOTit=∑j=1NGDPjt×dij−1, (1) with dij the great circle distance in kilometer between the centroids of provinces i and j.3 In practice, this indicator equates the potential demand for goods and services produced in a given location with that location’s proximity to consumer markets. Thus, it can be interpreted as the volume of economic activity to which a region has access to, after having taken into account the necessary transport costs to cover the distance to reach other provinces. An important point is that historical GDP estimates at the provincial level for the case of Italy are not available. Thus, we proxy for provincial GDP by allocating total regional GDP (NUTS-2 units) estimates for 1871, 1881, 1901 and 1911 to provinces (NUTS-3 units) using the provincial shares of regional population obtained by population census (see the Appendix for further details). Figure 6 shows that the values of Harris market potential are increasingly concentrated in north-western provinces. However, estimated market potential is also high in Florence, Rome, Naples and Palermo, that is the provinces with the pre-unitarian city capitals of, respectively, the Grand Duchy of Tuscany, the Papal States and the Kingdom of the Two Sicilies. This last evidence fits perfectly well with the intuition of Fenoaltea (2003, 1073–74) who, in his study on Italian regional industrialization, noticed that in the early 1870s ‘The industrial, manufacturing regions are those with the former capitals, of the preceding decades and centuries’ and that ‘In such a context the appropriate unit of analysis is not in fact the region, but (in Italy) the much smaller province’. Figure 6 View largeDownload slide Harris market potential. Figure 6 View largeDownload slide Harris market potential. 3.3. Expected sectoral effects of market potential and literacy The international literature, as we briefly summarized at the beginning of this section, shows that both market potential and factor endowment contribute to explain industrial location patterns in Britain, Poland, Spain and in the USA during the late 19th and early 20th centuries. However, the effects are generally sector-specific. Recall for instance that both Crafts and Mulatu (2006) for the UK and Klein and Crafts (2012) for the USA show that market potential was an attraction especially for industries with large plant size. This finding is important in that it supports the view that the effect of market potential may depend on the degree of scale economies prevailing in the various sectors. Effectively, economic geography theory predicts that economic activities with increasing returns to scale tend to establish themselves in regions that enjoy good market access, while the location of economic activities with constant returns technologies is mainly influenced by factor endowment. Sectoral differences may also exist in the effect of regional skill endowment on industrial location. Standard neoclassical trade theory (Heckscher–Ohlin model) predicts that high-skilled labor intensive industries tend to be concentrated in regions with higher endowment of high-skilled labor. Crafts and Mulatu (2006) show that in 19th-century Britain educated workers were an important input for the chemical sector, but not for textiles. Based on these economic geography arguments, one may broadly expect that the location of sectors characterized by high and medium capital to labor ratio (K/L), inherently tied to skill intensity and increasing returns to scale, was relatively more influenced by market potential. Table 2 groups the Italian manufacturing sectors into light and heavy industries depending on their capital intensity (the Appendix reports maps illustrating the spatial distribution of selected high and low K/L sectors). The table reports in particular data on horse-power per worker (HP/L), often used to proxy capital intensity (e.g., Broadberry and Crafts, 1990). The table is mostly based on the data from the first industrial census of 1911 given in Zamagni (1978), and the 0.6 for 1911 threshold for ‘K/L’ inferred from Federico (2006).4 It also uses the disaggregated data for the engineering sector by Fenoaltea (2016). There is little argument in the literature that metalmaking, paper and chemicals represent high ‘K/L’ sectors, while non-metallic mineral products, wood, leather and clothing are low ‘K/L’ sectors. Within the textile industry, cotton and wool belong to the high ‘K/L’ group, while silk belongs to the low one. The production of raw silk, an activity at the boundary between agriculture and manufacturing, is a very labor-intensive activity, suitable for water-rich and densely populated areas, such as the North-West of Italy. In addition, silk represented a leading component of Italian exports toward North-Western Europe, and being a luxury good, beyond the reach of most of 19th-century Italians (Federico and Tena-Junguito, 2014). As far as the engineering industry is concerned, Fenoaltea (2016) shows that the average value of K/L = 0.33 for the whole industry actually includes values as low as 0.20 for blacksmiths and as high as 0.60 for shipyards, reflecting the fact that the manufacturing of major naval vessels (but also steam locomotives) was far more sophisticated from a technological point of view than the maintenance of agricultural tools by blacksmiths.5 A final note on the foodstuffs sector is that, as warned by Zamagni (1978), the relatively high level of horse-power per worker should not be misinterpreted, since it was largely due to the traditional flour-milling industry. Table 2 Horse-power per worker (HP/L) in 1911 High HP/L Metalmaking 2.62 Chemicals and rubber 1.30 Foodstuffsa 0.94 Textiles     Cotton 0.85     Wool 0.78 Paper 0.73 Engineering     Machinery 0.61     Shipbuilding 0.60 Low HP/L Non-metallic mineral products 0.36 Wood 0.23 Engineering     Blacksmiths 0.20 Textiles     Silk 0.11 Leather 0.09 Clothing 0.07 High HP/L Metalmaking 2.62 Chemicals and rubber 1.30 Foodstuffsa 0.94 Textiles     Cotton 0.85     Wool 0.78 Paper 0.73 Engineering     Machinery 0.61     Shipbuilding 0.60 Low HP/L Non-metallic mineral products 0.36 Wood 0.23 Engineering     Blacksmiths 0.20 Textiles     Silk 0.11 Leather 0.09 Clothing 0.07 aFoodstuffs is net of sugar (with K/L = 2.18). Source: Zamagni (1978) and Fenoaltea (2016) for machinery, shipbuilding and blacksmiths. The ‘K/L’ figure for machinery is in particular the average of rail-guided vehicles, other heavy equipment and other ordinary machinery. Table 2 Horse-power per worker (HP/L) in 1911 High HP/L Metalmaking 2.62 Chemicals and rubber 1.30 Foodstuffsa 0.94 Textiles     Cotton 0.85     Wool 0.78 Paper 0.73 Engineering     Machinery 0.61     Shipbuilding 0.60 Low HP/L Non-metallic mineral products 0.36 Wood 0.23 Engineering     Blacksmiths 0.20 Textiles     Silk 0.11 Leather 0.09 Clothing 0.07 High HP/L Metalmaking 2.62 Chemicals and rubber 1.30 Foodstuffsa 0.94 Textiles     Cotton 0.85     Wool 0.78 Paper 0.73 Engineering     Machinery 0.61     Shipbuilding 0.60 Low HP/L Non-metallic mineral products 0.36 Wood 0.23 Engineering     Blacksmiths 0.20 Textiles     Silk 0.11 Leather 0.09 Clothing 0.07 aFoodstuffs is net of sugar (with K/L = 2.18). Source: Zamagni (1978) and Fenoaltea (2016) for machinery, shipbuilding and blacksmiths. The ‘K/L’ figure for machinery is in particular the average of rail-guided vehicles, other heavy equipment and other ordinary machinery. 4. Econometric analysis This section examines the relevance of factor endowment and (domestic) market potential in shaping the location of industries across Italian provinces during the period 1871–1911. Industrial location is measured in relative terms, that is by using the log of the location quotient, ln⁡(LQ). Market potential is measured by the log of Harris (1954) formula ( ln⁡(Mktpot)). As for factor endowment, we focus on labor skills and water abundance. Labor skills are measured by the log of literacy, ln⁡(Literacy), that is, the share of people aged 6 years and above who are able to read and write. Water supply is measured by the continuous, but time-invariant, variable ln⁡(River), and its interaction with the dummy variable Alpine, indicating if the province belongs to the Alpine region. Following Combes and Gobillon (2015), we test the effect of market potential, literacy and natural advantages in two steps. In the first step, we exploit the panel structure of the data (69 provinces for four time periods) to assess the effect of time-varying variables (i.e., ln⁡(Mktpot) and ln⁡(Literacy)), while in the second step we estimate the effect of time-invariant variables (i.e., water abundance). 4.1. The effect of market potential and literacy 4.1.1. Empirical strategy In this section, we discuss the estimated effects of the two time-varying variables (i.e., ln⁡(Mktpot) and ln⁡(Literacy)). Differently from Combes and Gobillon (2015), however, in this first step we control for time-invariant unobserved heterogeneity by using a simple semiparametric model with a smooth spatial trend (the so-called Geoadditive Model) (Lee and Durbán, 2011), rather than by introducing spatial-fixed effects. More formally, for each sector k, and denoting with i and t the province and time index, the model for the first step is specified as6: ln⁡(LQi,t)=α+β1ln⁡(Literacyi,t)+β2ln⁡(Mktpoti,t)+f1(Lati)+f2(Longi)+f12(Lati,Longi)+γt+εi,tεi,t ∼ iidN(0,σε2) i=1,…,N t=1,…,T. (2) Time-fixed effects (γt) are introduced in the model to control for time-related factor biases. Moreover, the geoadditive terms, that is, the smooth effect of the latitude— f1(Lati), of the longitude— f2(Longi), and of their interaction— f12(Lati,Longi)—work as control functions (CFs) to filter the spatial trend out of the residuals, and transfer it to the mean response in a model specification. Thus, they allow to capture the shape of the spatial distribution of the dependent variable, conditional on the determinants included in the model. These CFs also isolate stochastic spatial dependence in the residuals that is spatially autocorrelated unobserved heterogeneity (see also Basile et al., 2014). Thus, they can be regarded as an alternative to individual regional dummies (spatial-fixed effects) to capture unobserved spatial heterogeneity as long as the latter is smoothly distributed over space. Regional dummies peak significantly higher and lower levels of the mean response variable. If these peaks are smoothly distributed over a two-dimensional surface (i.e., if unobserved spatial heterogeneity is spatially auto-correlated), the smooth spatial trend is able to capture them.7 We simply demonstrate the validity of these statements by estimating the two competing models without explanatory variables (see the Appendix). A complication with the estimation of model (2) is given by the presence of endogenous variables— ln⁡(Mktpoti,t) and ln⁡(Literacyi,t)—on the right-hand side (r.h.s.). As for ln⁡(Mktpoti,t), NEG models describe a process characterized by reverse causality in which market potential, by attracting firms and workers, increases production in a particular location, and this, in turn, raises its market potential. ln⁡(Literacyi,t) may also be an endogenous variable. On the one hand, the availability of literate workers may foster the concentration of industrial activities in certain regions. On the other hand, however, more industrialized regions may provide an incentive to achieve education that is generally lacking in backward areas of the country. To address these issues, we extend the REML methodology to estimate the parameters of model (2) in a 2-stage ‘CF’ approach (Blundell and Powell, 2003), that is an alternative to standard instrumental variable/two-stage least square (IV–2SLS) methods. In the first stage, each endogenous variable is regressed on a set of conformable IVs (Z), using a semiparametric model. The residuals from the first stages are then included in the original model (2) to control for the endogeneity of ln⁡Mktpoti,t and ln⁡Literacyi,t. Since the second stage contains generated regressors (i.e., the first-step residuals), a bootstrap procedure is used to compute p-values [see Basile et al. (2014) for details on the bootstrap procedure]. This procedure requires finding good instruments, that is, variables that are correlated with the endogenous explanatory variables but not with the residuals of the regression. To control for the endogeneity of market potential, we follow the main empirical literature in using a measure of centrality of the region ( Centrality=∑idij−1) (Head and Mayer, 2006), and the geographical distance from the main economic center (i.e., the distance from Milan, DistMilani) (Redding and Venables, 2004; Wolf, 2007; Klein and Crafts, 2012; Martinez-Galarraga, 2012) as IVs. To control for the endogeneity of ln⁡Literacyi,t, we use its time lag ( ln⁡Literacyi,t−10). 4.1.2. Evidence for the whole manufacturing sector For the case of the whole manufacturing activity, we report in Table 3 the estimation results of the semiparametric CF approach, along with the estimation results of a fully parametric 2SLS. Obviously, we cannot use the within-group version of the 2SLS estimator, since two important instruments (DistMilani and Centralityi) are time invariant, while the third one ( ln⁡Literacyi,t−10) would be correlated with the within-group transformed error term. Thus, in order to control for spatial heterogeneity, we include a parametric nonlinear spatial trend (i.e., Lat, Lat2, Long, Long2, Lat × Long) on the r.h.s. of the pooled 2SLS model. Table 3 Whole manufacturing. Estimation results of the parametric IV (2SLS) approach and of the semiparametric control function (CF) approach Variable 2SLS Semiparametric CF Parametric terms     Intercept 5.003 −1.833***      (0.613) (0.001)      ln⁡(Mktpot) 0.610*** 0.468***      (0.000) (0.000)      ln⁡(Literacy) 0.015 0.235**      (0.934) (0.027)     Lat −0.487      (0.238)     Lat2 0.006      (0.153)     Long 0.268      (0.456)     Long2 −0.009**      (0.027)     Lat×Long −0.001      (0.940) Non-parametric terms      f1(Lat) 7.760***      (0.000)      f2(Long) 10.259***      (0.000)      f12(Lat,Long) 20.047***      (0.000)      h1(Res1) 3.007***      (0.078)      h2(Res2) 2.933***      (0.002) Diagnostics     Weak instr.-ln(Mktpot) 21.208*** 24.461*      (0.000) (0.000)     Weak instr.-ln(Literacy) 405.489*** 210.072***      (0.000) (0.000)     Wu–Hausman 10.535*** 33.056***      (0.000) (0.000)     Sargan 0.436 3.727      (0.509) (0.155) Variable 2SLS Semiparametric CF Parametric terms     Intercept 5.003 −1.833***      (0.613) (0.001)      ln⁡(Mktpot) 0.610*** 0.468***      (0.000) (0.000)      ln⁡(Literacy) 0.015 0.235**      (0.934) (0.027)     Lat −0.487      (0.238)     Lat2 0.006      (0.153)     Long 0.268      (0.456)     Long2 −0.009**      (0.027)     Lat×Long −0.001      (0.940) Non-parametric terms      f1(Lat) 7.760***      (0.000)      f2(Long) 10.259***      (0.000)      f12(Lat,Long) 20.047***      (0.000)      h1(Res1) 3.007***      (0.078)      h2(Res2) 2.933***      (0.002) Diagnostics     Weak instr.-ln(Mktpot) 21.208*** 24.461*      (0.000) (0.000)     Weak instr.-ln(Literacy) 405.489*** 210.072***      (0.000) (0.000)     Wu–Hausman 10.535*** 33.056***      (0.000) (0.000)     Sargan 0.436 3.727      (0.509) (0.155) Notes: Coefficients, e.d.f. and bootstrap p-values (in parenthesis). Time-fixed effects are included in both models. Number of observations: 276. Table 3 Whole manufacturing. Estimation results of the parametric IV (2SLS) approach and of the semiparametric control function (CF) approach Variable 2SLS Semiparametric CF Parametric terms     Intercept 5.003 −1.833***      (0.613) (0.001)      ln⁡(Mktpot) 0.610*** 0.468***      (0.000) (0.000)      ln⁡(Literacy) 0.015 0.235**      (0.934) (0.027)     Lat −0.487      (0.238)     Lat2 0.006      (0.153)     Long 0.268      (0.456)     Long2 −0.009**      (0.027)     Lat×Long −0.001      (0.940) Non-parametric terms      f1(Lat) 7.760***      (0.000)      f2(Long) 10.259***      (0.000)      f12(Lat,Long) 20.047***      (0.000)      h1(Res1) 3.007***      (0.078)      h2(Res2) 2.933***      (0.002) Diagnostics     Weak instr.-ln(Mktpot) 21.208*** 24.461*      (0.000) (0.000)     Weak instr.-ln(Literacy) 405.489*** 210.072***      (0.000) (0.000)     Wu–Hausman 10.535*** 33.056***      (0.000) (0.000)     Sargan 0.436 3.727      (0.509) (0.155) Variable 2SLS Semiparametric CF Parametric terms     Intercept 5.003 −1.833***      (0.613) (0.001)      ln⁡(Mktpot) 0.610*** 0.468***      (0.000) (0.000)      ln⁡(Literacy) 0.015 0.235**      (0.934) (0.027)     Lat −0.487      (0.238)     Lat2 0.006      (0.153)     Long 0.268      (0.456)     Long2 −0.009**      (0.027)     Lat×Long −0.001      (0.940) Non-parametric terms      f1(Lat) 7.760***      (0.000)      f2(Long) 10.259***      (0.000)      f12(Lat,Long) 20.047***      (0.000)      h1(Res1) 3.007***      (0.078)      h2(Res2) 2.933***      (0.002) Diagnostics     Weak instr.-ln(Mktpot) 21.208*** 24.461*      (0.000) (0.000)     Weak instr.-ln(Literacy) 405.489*** 210.072***      (0.000) (0.000)     Wu–Hausman 10.535*** 33.056***      (0.000) (0.000)     Sargan 0.436 3.727      (0.509) (0.155) Notes: Coefficients, e.d.f. and bootstrap p-values (in parenthesis). Time-fixed effects are included in both models. Number of observations: 276. All in all, the diagnostic tests of the 2SLS model provide encouraging evidence in favor of the chosen set of instruments. First, the Wu–Hausman test confirms that ln⁡Literacyi,t and ln⁡Mktpoti,t are endogenous. Second, the weak instrument tests confirm that the IVs are strongly correlated to the endogenous variables. Third, the Sargan test of overidentifying restrictions suggests that the DistMilani, Centralityi, and ln⁡Literacyi,t−10 are valid instruments, that is, they are uncorrelated with the error term, and thus they are correctly excluded from the estimated equation. Nevertheless, the estimated parametric 2SLS model does not properly control for unobserved spatial heterogeneity. From Table 3 it emerges indeed that the spatial variables (Lat, Lat2, Long, Long2, Lat × Long) are weakly significant. The spatial trend in the data (i.e., the smooth spatial heterogeneity) is much better captured by the semiparametric geoadditive model as indicated by the high significance of the smooth terms f1, f2, and f12 (in column semiparametric CF).8 The smooth functions of the residuals from the two first stages ( h1(Res1) and h2(Res2)) work as CFs to correct the estimated parameters for the endogeneity bias. The statistical significance of these smooth terms can also be used as endogeneity tests. In Table 3 we also report a test of the joint significance of h1(Res1) and h2(Res2), similar to the Wu–Hausman test in the parametric 2SLS approach, as well as two weak-instrument tests using the results of the semiparametric first stages (which confirm that the excluded instruments are strongly correlated to the endogenous variables, thus rejecting the hypothesis of weak instruments). Unfortunately, there is not a well-known and widely accepted test for the validity of the conditional mean restrictions imposed by the CF approach (overidentification test). A practically feasible way of testing such restrictions consists of including the excluded instruments in the CF estimate, and testing their significance. They should not be significant since the CF should pick up all of the correlation between the structural error term and (X, Z). In our case, all the external instruments turned out to be strictly exogenous. The CF estimation results for the whole manufacturing industry show that, after controlling for unobserved heterogeneity and after correcting for the endogeneity bias, the variables ln⁡(Literacy) and ln⁡(Mktpot) enter the model significantly and positively, thus proving to be important drivers of industrial location during the sample period. In particular, the positive effect of market potential corroborates the hypothesis that even during the early stage of Italy’s industrialization firms tended to settle in regions with the highest market potential (i.e., in the North West). Moreover, the elasticity of Mktpot (0.47) is much higher than that of Literacy (0.24). 4.1.3. The heterogeneous effect of market potential and literacy across sectors Table 4 looks within manufacturing and provides the estimation results of model (2) for each industrial sector. Again, the model includes a smooth spatial trend to control for unobserved spatial heterogeneity, time dummies to control for unobserved time heterogeneity and smooth terms of first-stage residuals as CFs to correct the inconsistency due to the endogeneity of ln⁡(Mktpot) and ln⁡(Literacy).9 Table 4 Marginal effects of ln(Mktpot) and ln(Literacy) Sectors Market potential Literacy Whole manufacturing 0.468*** 0.235**      (0.000) (0.027) High HP/L sectors     Metalmaking 1.502*** 1.976***      (0.000) (0.002)     Chemicals and rubber 0.653*** 0.670**      (0.000) (0.044)     Foodstuffs −0.286*** 0.272***      (0.000) (0.002)     Cotton 2.815*** −0.244      (0.000) (0.817)     Wool 0.775* −1.475*      (0.073) (0.058)     Other natural fibers 0.614** 0.220      (0.018) (0.701)     Paper 0.969*** 0.320      (0.000) (0.217)     Machinery 0.687*** 0.105      (0.000) (0.629)     Shipbuilding 0.475 2.659**      (0.469) (0.015) Low HP/L sectors     Non-metallic mineral products 0.023 −0.770***      (0.862) (0.001)     Wood −0.191*** 0.199**      (0.001) (0.019)     Blacksmith −0.222*** −0.184*      (0.000) (0.054)     Silk 2.199*** −1.035      (0.000) (0.321)     Leather −0.453*** −0.005      (0.000) (0.963)     Clothing 0.017 0.126      (0.876) (0.406) Sectors Market potential Literacy Whole manufacturing 0.468*** 0.235**      (0.000) (0.027) High HP/L sectors     Metalmaking 1.502*** 1.976***      (0.000) (0.002)     Chemicals and rubber 0.653*** 0.670**      (0.000) (0.044)     Foodstuffs −0.286*** 0.272***      (0.000) (0.002)     Cotton 2.815*** −0.244      (0.000) (0.817)     Wool 0.775* −1.475*      (0.073) (0.058)     Other natural fibers 0.614** 0.220      (0.018) (0.701)     Paper 0.969*** 0.320      (0.000) (0.217)     Machinery 0.687*** 0.105      (0.000) (0.629)     Shipbuilding 0.475 2.659**      (0.469) (0.015) Low HP/L sectors     Non-metallic mineral products 0.023 −0.770***      (0.862) (0.001)     Wood −0.191*** 0.199**      (0.001) (0.019)     Blacksmith −0.222*** −0.184*      (0.000) (0.054)     Silk 2.199*** −1.035      (0.000) (0.321)     Leather −0.453*** −0.005      (0.000) (0.963)     Clothing 0.017 0.126      (0.876) (0.406) Notes: Coefficients and bootstrap p-values (in parenthesis). The number of observations for each sector is 276 (69 province by four time points). Table 4 Marginal effects of ln(Mktpot) and ln(Literacy) Sectors Market potential Literacy Whole manufacturing 0.468*** 0.235**      (0.000) (0.027) High HP/L sectors     Metalmaking 1.502*** 1.976***      (0.000) (0.002)     Chemicals and rubber 0.653*** 0.670**      (0.000) (0.044)     Foodstuffs −0.286*** 0.272***      (0.000) (0.002)     Cotton 2.815*** −0.244      (0.000) (0.817)     Wool 0.775* −1.475*      (0.073) (0.058)     Other natural fibers 0.614** 0.220      (0.018) (0.701)     Paper 0.969*** 0.320      (0.000) (0.217)     Machinery 0.687*** 0.105      (0.000) (0.629)     Shipbuilding 0.475 2.659**      (0.469) (0.015) Low HP/L sectors     Non-metallic mineral products 0.023 −0.770***      (0.862) (0.001)     Wood −0.191*** 0.199**      (0.001) (0.019)     Blacksmith −0.222*** −0.184*      (0.000) (0.054)     Silk 2.199*** −1.035      (0.000) (0.321)     Leather −0.453*** −0.005      (0.000) (0.963)     Clothing 0.017 0.126      (0.876) (0.406) Sectors Market potential Literacy Whole manufacturing 0.468*** 0.235**      (0.000) (0.027) High HP/L sectors     Metalmaking 1.502*** 1.976***      (0.000) (0.002)     Chemicals and rubber 0.653*** 0.670**      (0.000) (0.044)     Foodstuffs −0.286*** 0.272***      (0.000) (0.002)     Cotton 2.815*** −0.244      (0.000) (0.817)     Wool 0.775* −1.475*      (0.073) (0.058)     Other natural fibers 0.614** 0.220      (0.018) (0.701)     Paper 0.969*** 0.320      (0.000) (0.217)     Machinery 0.687*** 0.105      (0.000) (0.629)     Shipbuilding 0.475 2.659**      (0.469) (0.015) Low HP/L sectors     Non-metallic mineral products 0.023 −0.770***      (0.862) (0.001)     Wood −0.191*** 0.199**      (0.001) (0.019)     Blacksmith −0.222*** −0.184*      (0.000) (0.054)     Silk 2.199*** −1.035      (0.000) (0.321)     Leather −0.453*** −0.005      (0.000) (0.963)     Clothing 0.017 0.126      (0.876) (0.406) Notes: Coefficients and bootstrap p-values (in parenthesis). The number of observations for each sector is 276 (69 province by four time points). The results were obtained by pooling the data over the 69 provinces and the four census years (1871, 1881, 1901 and 1911). For each parametric term, we report the estimated coefficient and the corresponding bootstrapped p-value. In line with our expectations, during the period 1871–1911 market potential turned out to be a key driver of the industrial location in the case of high HP/L industries. The coefficient associated to the variable ln⁡(Mktpot) is indeed positive and strongly significant in the case of metalmaking, chemicals, paper, machinery, cotton, wool and other natural fibers. The elasticity of this variable is particularly high in the case of cotton (2.8) and metalmaking (1.5). There are, however, two exceptions. The first is shipbuilding, for which the effect of market potential is not statistically significant. But in this case, the industrial location was obviously driven by the presence of a port, independently of the market potential of the region. The second exception is foodstuffs, for which the effect is negative and significant. Census data for 1911 inform us that some 50% and more of this industrial activity was accounted for by the first- and second-stage processing of wheat and other cereals (flour-milling industry, and the baking of bread, biscuits and pasta production). These activities need often to locate close to cities, regardless of their position in the urban hierarchy. Moreover, the mild climate of the South of Italy represented a strong comparative advantage for the production of some foodstuffs, such as pasta asciutta (literally ‘dry pasta’) traditionally desiccated through simple exposition to the air. The effect of ln⁡(Mktpot) was instead negative (or not significant) in the case of low HP/L industries. Silk represents an isolated, yet important, exception. It was a labor intensive sector for which the estimated effect of market potential is sizeable (2.2), and highly significant. Recall, however, that silk production was increasingly concentrated in northern provinces and largely exported to northern European countries. Thus, it is likely that proximity to northern Europe and decreasing international physical transport costs associated with railway development, more than domestic market potential, were at the heart of the relocation of the silk industry to the northern Italian provinces.10 These findings are consistent with the predictions of the Core–Periphery NEG model (Krugman, 1991), according to which economic activities with increasing returns to scale tend to establish themselves in regions that enjoy good market access. A region with a larger market potential is also characterized by a more generous compensation of local factors, while small regions that are far from the large market will have lower local wages compared to regions close to the industrial core. Thus, if everything else is unchanged, and if the firms all achieve constant returns to scale (as in the case of low HP/L industries), the increase in the price of local factors (labor, land and so on) will reduce the profitability of all the firms at that location. Within high HP/L sectors, shipbuildings, metalmaking, foodstuffs and chemicals proved to be knowledge-intensive sectors requiring educated labor force, mainly located in the north-western regions. While, for most traditional low HP/L sectors, often tied to agriculture, literacy has a negative impact or is not significant, with wood representing the only exception. 4.2. The effect of water endowment 4.2.1. Empirical strategy and evidence for the whole manufacturing sector Similarly to Combes and Gobillon (2015), in the second step of our empirical strategy we assess the effect of time-invariant variables (i.e., Alpine, ln River and their interaction). However, instead of regressing the estimated-fixed effects, αî, on Alpinei, ln⁡(Riveri) and Alpinei×ln⁡(Riveri), we regress the estimated values of the spatial trend ( fspt̂i) on these three variables. That is, we estimate the following interaction model using OLS: fspt̂i=α+β1Alpinei+β2ln⁡(Riveri)++β3Alpinei×ln⁡(Riveri)+εiεi∼iidN(0,σε2)i=1,…,Nfspt̂i=f1̂(Lati)+f2̂(Longi)+f12̂(Lati,Longi). (3) As expected, however, the OLS residuals turned out to be spatially autocorrelated. For the case of the whole manufacturing sector, the standardized Moran I test statistic on the residuals is equal to 4.539 and its p-value is 0.000. In fact, Alpinei, ln⁡(Riveri) and Alpinei×ln⁡(Riveri) capture only a portion of the large spatial heterogeneity which characterizes the spatial distribution of the location quotient net of the effect of market potential and literacy, that is, the spatial distribution of fspt̂i (the adjusted R2 of the model is 0.31). Heterogeneity among provinces, induced by an uneven distribution of immobile resources (such as natural harbors) and amenities (climate) may also be at the origin of a variety of comparative advantages. This unobserved heterogeneity may also be spatially correlated, thus introducing spatial autocorrelation in the residuals. In order to control for spatial autocorrelation, we use a semiparametric spatial filter (Tiefelsdorf and Griffith, 2007). As is well known, spatial filtering uses a set of spatial proxy variables, which are extracted as eigenvectors from the spatial weights matrix, and implants these vectors as control variables into the model. These control variables identify and isolate the stochastic spatial dependencies among observations, thus allowing model building to proceed as if the observations were independent. Specifically, we used the function SpatialFiltering from the R library spdep. This function selects eigenvectors in a semiparametric spatial filtering approach (as proposed by Tiefelsdorf and Griffith, 2007) to remove spatial dependence from linear models. The optimal subset of eigenvectors is identified by an objective function that minimizes spatial autocorrelation, that is, by finding the single eigenvector which reduces the standard variate of Moran’s I for regression residuals most, and continuing until no candidate eigenvector reduces the value by more than a tolerance value. This subset of eigenvectors is a proxy either for those spatially autocorrelated exogenous factors that have not been incorporated into a model, or for an underlying spatial process that ties the observations together. Furthermore, incorporation of all relevant eigenvectors into a model should leave the remaining residual component spatially uncorrelated. Consequently, standard statistical modeling and estimation techniques as well as interpretations can be employed for spatially filtered models. The spatially filtered OLS estimates of model (3) confirm the widely accepted idea that, over the analyzed period, the whole manufacturing activity was mainly located in proximity to Alpine rivers. Figure 7 displays in panel (a) the changes in the marginal effect of Alpine conditional on the values of ln⁡River, that is ∂fspt̂̂∂Alpine=β1+β3ln⁡River, while it shows in panel (b) the changes in the marginal effect of the continuous variable ln⁡River conditional on the values of the dummy variable Alpine, that is ∂fspt̂̂∂ln⁡River=β2+β3Alpine. Figure 7 View largeDownload slide Whole manufacturing. Marginal effect of ln(River) and Alpine with simulated 95% confidence intervals. (a) Estimated coefficient of Alpine by ln(River) and (b) Estimated coefficient of ln(River) by Alpine. Figure 7 View largeDownload slide Whole manufacturing. Marginal effect of ln(River) and Alpine with simulated 95% confidence intervals. (a) Estimated coefficient of Alpine by ln(River) and (b) Estimated coefficient of ln(River) by Alpine. The two plots, created using the R function interplot, also include simulated 95% pointwise confidence intervals (obtained using the simulation function from the R package arm of Gelman and Hill, 2006) around these marginal effects. The plot in panel (a) shows that, with increasing importance of the river (along the horizontal axis), the magnitude of the marginal effect of Alpine on the location of the whole manufacturing sector also increases (along the vertical axis). The ‘dot-and-whisker’ plot in panel (b) shows the effect of ln⁡River (measuring the importance of the river) conditional on the dummy variable Alpine. On the one hand, when Alpine = 1 (i.e., when the province belongs to the Alpine region), the simulated 95% confidence interval around the marginal effect of ln⁡River is above (and does not contain) the horizontal zero line. On the other hand, when Alpine = 0 (i.e., when the province does not belong to the Alpine region), the simulated 95% confidence interval around the marginal effect of ln⁡River is below (and does not contain) the horizontal zero line. Thus, for manufacturing, there is a positive effect of Alpine rivers, as widely suggested by previous qualitative studies of historical nature stressing the importance of water as a source of motive power and, more generally, to sustain manufacturing activities that required water throughout the year. 4.2.2. The heterogeneous effect of water endowment across sectors Table 5 shows that, once the effect of market potential and literacy is accounted for, the marginal effect of ln⁡River when Alpine = 1 is positive and statistically significant for 6 out of 15 industries. They are evenly shared between low HP/L industries (clothing, silk and leather) and high HP/L sector (chemicals, foodstuffs and other natural fibers). In the case of chemical products, we find in particular the same result holding for the whole manufacturing industry (i.e., a positive marginal effect of Alpine rivers and a negative effect of Apennine rivers). The estimated effect is instead negative or insignificant for the remaining industries. Apennine rivers have instead a positive effect in the cases of wool, silk, blacksmith and leather. Table 5 Marginal effect of ln(River) conditional on Alpine Sectors Alpine = 0 Lower bound Upper bound Alpine = 1 Lower bound Upper bound Manufacturing −0.208* −0.281 −0.133 0.127* 0.013 0.241 High HP/L sectors     Metalmaking −0.294* −0.532 −0.059 −0.283 −0.632 0.074     Chemicals and rubber −0.198* −0.333 −0.064 0.430* 0.221 0.646     Foodstuffs 0.055 −0.018 0.130 0.150* 0.041 0.257     Cotton −0.200 −0.459 0.049 −1.023* −1.389 −0.656     Wool 0.712* 0.256 1.162 −0.002 −0.679 0.681     Other natural fibers 0.029 −0.130 0.189 0.528* 0.292 0.767     Paper −0.095 −0.215 0.025 −0.225* −0.403 −0.052     Machinery −0.040 −0.098 0.015 0.031 −0.054 0.114     Shipbuilding −1.460* −2.254 −0.661 −0.557 −1.730 0.643 Low HP/L sectors     Non-metallic mineral products 0.045 −0.074 0.164 −0.397* −0.589 −0.204     Wood −0.026 −0.091 0.039 0.051 −0.051 0.152     Blacksmith 0.089* 0.046 0.132 −0.033 −0.101 0.035     Silk 0.492* 0.016 0.961 0.731* 0.034 1.452     Leather 0.075* 0.009 0.141 0.156* 0.062 0.250     Clothing 0.078 −0.036 0.188 0.189* 0.018 0.361 Sectors Alpine = 0 Lower bound Upper bound Alpine = 1 Lower bound Upper bound Manufacturing −0.208* −0.281 −0.133 0.127* 0.013 0.241 High HP/L sectors     Metalmaking −0.294* −0.532 −0.059 −0.283 −0.632 0.074     Chemicals and rubber −0.198* −0.333 −0.064 0.430* 0.221 0.646     Foodstuffs 0.055 −0.018 0.130 0.150* 0.041 0.257     Cotton −0.200 −0.459 0.049 −1.023* −1.389 −0.656     Wool 0.712* 0.256 1.162 −0.002 −0.679 0.681     Other natural fibers 0.029 −0.130 0.189 0.528* 0.292 0.767     Paper −0.095 −0.215 0.025 −0.225* −0.403 −0.052     Machinery −0.040 −0.098 0.015 0.031 −0.054 0.114     Shipbuilding −1.460* −2.254 −0.661 −0.557 −1.730 0.643 Low HP/L sectors     Non-metallic mineral products 0.045 −0.074 0.164 −0.397* −0.589 −0.204     Wood −0.026 −0.091 0.039 0.051 −0.051 0.152     Blacksmith 0.089* 0.046 0.132 −0.033 −0.101 0.035     Silk 0.492* 0.016 0.961 0.731* 0.034 1.452     Leather 0.075* 0.009 0.141 0.156* 0.062 0.250     Clothing 0.078 −0.036 0.188 0.189* 0.018 0.361 Notes: Coefficients and bootstrap confidence intervals. The number of observations for each sector is 69. Table 5 Marginal effect of ln(River) conditional on Alpine Sectors Alpine = 0 Lower bound Upper bound Alpine = 1 Lower bound Upper bound Manufacturing −0.208* −0.281 −0.133 0.127* 0.013 0.241 High HP/L sectors     Metalmaking −0.294* −0.532 −0.059 −0.283 −0.632 0.074     Chemicals and rubber −0.198* −0.333 −0.064 0.430* 0.221 0.646     Foodstuffs 0.055 −0.018 0.130 0.150* 0.041 0.257     Cotton −0.200 −0.459 0.049 −1.023* −1.389 −0.656     Wool 0.712* 0.256 1.162 −0.002 −0.679 0.681     Other natural fibers 0.029 −0.130 0.189 0.528* 0.292 0.767     Paper −0.095 −0.215 0.025 −0.225* −0.403 −0.052     Machinery −0.040 −0.098 0.015 0.031 −0.054 0.114     Shipbuilding −1.460* −2.254 −0.661 −0.557 −1.730 0.643 Low HP/L sectors     Non-metallic mineral products 0.045 −0.074 0.164 −0.397* −0.589 −0.204     Wood −0.026 −0.091 0.039 0.051 −0.051 0.152     Blacksmith 0.089* 0.046 0.132 −0.033 −0.101 0.035     Silk 0.492* 0.016 0.961 0.731* 0.034 1.452     Leather 0.075* 0.009 0.141 0.156* 0.062 0.250     Clothing 0.078 −0.036 0.188 0.189* 0.018 0.361 Sectors Alpine = 0 Lower bound Upper bound Alpine = 1 Lower bound Upper bound Manufacturing −0.208* −0.281 −0.133 0.127* 0.013 0.241 High HP/L sectors     Metalmaking −0.294* −0.532 −0.059 −0.283 −0.632 0.074     Chemicals and rubber −0.198* −0.333 −0.064 0.430* 0.221 0.646     Foodstuffs 0.055 −0.018 0.130 0.150* 0.041 0.257     Cotton −0.200 −0.459 0.049 −1.023* −1.389 −0.656     Wool 0.712* 0.256 1.162 −0.002 −0.679 0.681     Other natural fibers 0.029 −0.130 0.189 0.528* 0.292 0.767     Paper −0.095 −0.215 0.025 −0.225* −0.403 −0.052     Machinery −0.040 −0.098 0.015 0.031 −0.054 0.114     Shipbuilding −1.460* −2.254 −0.661 −0.557 −1.730 0.643 Low HP/L sectors     Non-metallic mineral products 0.045 −0.074 0.164 −0.397* −0.589 −0.204     Wood −0.026 −0.091 0.039 0.051 −0.051 0.152     Blacksmith 0.089* 0.046 0.132 −0.033 −0.101 0.035     Silk 0.492* 0.016 0.961 0.731* 0.034 1.452     Leather 0.075* 0.009 0.141 0.156* 0.062 0.250     Clothing 0.078 −0.036 0.188 0.189* 0.018 0.361 Notes: Coefficients and bootstrap confidence intervals. The number of observations for each sector is 69. Overall, the results shed a new light on the interpretation of the much debated role of water endowment on industrial location in 19th-century Italy. The historical literature stressed that during the early development of the Italian manufacturing industry, Alpine regions had a comparative advantage (over central and southern ones) stemming from water endowment. Effectively, even when the effect of market potential and literacy is duly accounted for, there is still room left for water endowment as a driver of industrial location. However, this role appears to be inherently tied to the nature of the industrial sector considered, and cannot simply be attributed to a generic Alpine/Apennine divide. Consider the case of the silk and leather industries. The descriptive analysis (reported in the Appendix) shows that silk was increasingly agglomerated in the Alpine region, while leather was spatially diffused in the South. Both industries made an intensive use of fresh water, not just as a source of motive power, but also and more generally in the various steps of their production process. For both industries, the empirical results of Table 5 show effectively that the proximity to a river was relevant independently of its position (either Alpine or Apennine). 5. Conclusions Using a new set of provincial (NUTS-3 units) data on industrial value added at 1911 prices, this paper analyzed the spatial location patterns of manufacturing activity in Italy during the period 1871–1911. Specifically, we tested the relative effect of tangible factors, such as market size and factor endowment (water abundance and literacy), as the main drivers of industrial location, ruling out the role of intangible factors such as knowledge spillovers. The results show that, after the political and economic unification of the country in 1861, Italian provinces became more and more specialized, and manufacturing activity became increasingly concentrated in a few provinces, mostly belonging to the North-West part of the country. The estimation results corroborate the hypothesis that both comparative advantages (water endowment effect) and market potential (home-market effect) have been responsible for this process of spatial concentration. These findings are in line with those available in the literature for other countries (see, e.g., Rosés (2003) for 19th-century Spain). The capital/labor (‘K/L’) ratio, related to both skill-intensity and increasing-returns-to scale, provided useful guidance in the analysis of sectoral developments. The role of market potential as a driver of industrial location emerged clearly for medium-high ‘K/L’ sectors mostly tied to the production of durable and capital goods (metalmaking, chemicals, machinery and paper), but also of consumer goods (cotton and wool). In addition, the location of metalmaking and chemicals was positively related to the availability of more educated labor force. Furthermore, once the effect of market potential and literacy is accounted for, we find evidence that the location of some traditional industries characterized by a low K/L ratio (such as silk and leather) was mainly driven by water endowment. There are of course more historical industrial location developments in heaven and earth than in our theoretical and empirical models. The case of shipbuilding, where only literacy emerged as an important driver of industrial location, does not fit well for instance within our interpretive scheme. With these caveats in mind, our results confirm the importance of using data that are disaggregated at the sectoral level when investigating the main factors influencing regional industrial location. Our findings are in line with the Core–Periphery NEG model (Krugman, 1991) that predicts increasing polarization and regional specialization as a result of economic integration. In this model, agglomeration economies derive from the interaction among economies of scale, transportation costs and market size, while intangible external economies (such as information spillovers) do not play any role. Today NEG models are strongly criticized on the basis of the observation that they focus on forces and processes that were important a century ago but are much less relevant today. The NEG approach seems less and less applicable to the actual location patterns of advanced economies. Nevertheless, the current economic geography of fast-growing countries like Brazil, China and India is highly reminiscent of the economic geography of Western countries during the 19th century, and it fits well into the NEG framework. Historical analyses like the one presented in this paper might thus help our understanding of the current process of economic modernization of developing countries. Footnotes 1 The increasing domestic market integration (decreasing trade costs) during the sample period can be documented through several evidences. First, in the early 1860s internal trade barriers were eliminated, and the mild external trade tariff previously adopted by the Kingdom of Savoy was extended to the rest of the country (Ciccarelli and Proietti, 2013). Second, internal transport costs were reduced through the development of the railway network: 8 km in 1839, 2400 km in 1860, 9100 km in 1880, and 15,300 km in 1910, when the network was essentially completed (ISTAT, 1958, 137). Finally, the reduction in shipping and railway rates during 1870–1910 is documented in Missiaia (201648). 2 Harris (1954)’s formula can be derived from a NEG theoretical model (Combes et al., 2008). An alternative way of obtaining a structural estimate of market potential based on NEG models was proposed by Redding and Venables (2004). The latter, however, requires data on bilateral trade flows that are not available at regional level for our sample period. 3 In order to measure Harris’ market potential, we rely on geodesic distances between locations. Recent historical studies (see e.g., Martínez-Galarraga, 2014) consider more sophisticated measures of bilateral transport costs. These measures take into consideration different transport modes, the routes used in the transportation of commodities by mode and the respective freight rates; they also take into account that their evolution over time can vary for a number of reasons. The adoption of this approach is beyond the scope of the present paper, since it requires a large amount of historical information that is not available using current data sources. Rather, it will be considered as an objective of our future research agenda. 4 Federico (2006) distinguishes between light industries and heavy industries. The former are those that ‘are featured by a (relatively) low capital intensity and by the prevailing orientation toward the final consumer’, the latter ‘are more capital-intensive and produce mainly inputs for other sectors’. The author (see in particular Table 2.6 therein), uses the industrial census of 1927 and considers as heavy industries those with a ‘K/L’ value above 0.7. Our data refer to 1911, and a 0.6 threshold for ‘K/L’ dovetails better with the data at hand. 5 The historical data and methodology used in Zamagni (1978) and Fenoaltea (2016) for the engineering sector differ, so that the average ‘K/L’ value for the year 1911 proposed by Zamagni (1978) (‘K/L’ = 0.51) and by Fenoaltea (2016) (‘K/L’ = 0.33) are difficult to compare. 6 This specification is different from the one used in the related literature (e.g., by Ellison and Glaeser, 1999; Midelfart-Knarvik et al., 2000; Wolf, 2007). These authors pool the data by regions, sectors and time, and regress the location quotient on a set of interactions between the vectors of location characteristics (factor endowment and market potential) and a vector of industry characteristics (measuring industries’ factor intensities and the share of intermediate inputs in GDP). Our dataset is rich enough to avoid considering the above interaction and pooling of the data, and it allows us to estimate separate models for each industrial sector. 7 The smooth components of the spatial trend in equation 2—f1, f2 and f12—are approximated by tensor product smoothers (Wood, 2006). As is well known, any semiparametric model can be expressed as a mixed model and, thus, it is possible to estimate all the parameters using restricted maximum-likelihood methods (REMLs). To estimate this model, we used the method described by Wood (2006) which allows for automatic and integrated smoothing parameters selection through the minimization of the REML. Wood has implemented this approach in the R package mgcv. 8 For the nonparametric smooth terms (i.e., the spatial trend and the first-stage residuals), Table 3 also shows the estimated degree of freedom (e.d.f.), a broad measure of nonlinearity (an e.d.f. equal to 1 indicates linearity, while a value higher than 1 indicates nonlinearity). 9 The results for these controls are not reported in Table 4, but are available upon request. 10 Up to the late 1880s, when Italy started a nonsense trade-war with France, nearly half of Italy’s total exports, albeit rather limited in size, were directed to the French market (Federico et al., 2011, 42–43). Acknowledgements We are grateful to Brian A'Hearn, Stefano Fachin, and Jacob Weisdorf for useful comments and suggestions. Research presented at the Berlin Colloquium of Economic History (Berlin, 2015), the Jean Paelinck Seminar of Spatial Econometric (Dijon 2015), the annual conferences of the European Economic History Society (EHES 2015, Pisa), Economic History Society (EHS 2015, Telford), the Italian Association of Regional Science (AISRe 2015, Rende), and the Italian Economic Association (SIE 2015, Naples). References A’Hearn B. , Venables A. ( 2013 ) Regional disparities: internal geography and external trade. In Toniolo G. (ed.) The Oxford Handbook of the Italian Economy since Unification . Oxford : Oxford University Press . Arbia G. , Dedominicis L. , De Groot H. ( 2013 ) The spatial distribution of economic activities in Italy. A sectorally detailed spatial data analysis using local labour market areas . Regional Studies , 47 : 405 – 418 . Google Scholar CrossRef Search ADS Basile R. , Durbán M. , Mínguez R. , Montero J. M. , Mur J. ( 2014 ) Modeling regional economic dynamics: spatial dependence, spatial heterogeneity and nonlinearities . Journal of Economic Dynamics and Control , 48 : 229 – 245 . Google Scholar CrossRef Search ADS Blundell R. , Powell J. ( 2003 ) Endogeneity in nonparametric and semiparametric regression models. In Dewatripont M. , Hansen L. , Turnsovsky S. J. (eds) Advances in Economics and Econometrics Theory and Application , pp. 312–357. Cambridge : Cambridge University Press . Broadberry S. N. , Crafts N. F. ( 1990 ) Explaining Anglo-American productivity differences in the mid-twentieth century . Oxford Bulletin of Economics and Statistics , 52 : 375 – 402 . Google Scholar CrossRef Search ADS Cafagna L. ( 1989 ) Dualismo e Sviluppo nella Storia d’Italia . Venice : Marsilio . Ciccarelli C. , Fenoaltea S. ( 2009 ) La Produzione Industriale delle Regioni d’Italia, 1861–1913: una Ricostruzione Quantitativa. 1. Le Industrie Non Manifatturiere . Rome : Bank of Italy . Ciccarelli C. , Fenoaltea S. ( 2013 ) Through the magnifying glass: provincial aspects of industrial growth in post-unification Italy . Economic History Review , 66 : 57 – 85 . Google Scholar CrossRef Search ADS Ciccarelli C. , Fenoaltea S. ( 2014 ) La Produzione Industriale delle Regioni d’Italia, 1861–1913: una Ricostruzione Quantitativa. 2. Le Industrie Estrattivo-Manifatturiere . Rome : Bank of Italy . Ciccarelli C. , Proietti T. ( 2013 ) Patterns of industrial specialisation in post-unification Italy . Scandinavian Economic History Review , 61 : 259 – 286 . Google Scholar CrossRef Search ADS Ciccarelli C. , Weisdorf J. L. ( 2016 ) The effect of the Italian unification on the comparative regional development in literacy, 1821–1911 . WP CEIS No. 392, University of Rome Tor Vergata . Combes P.-P. , Gobillon L. ( 2015 ) The empirics of agglomeration economies. In Gilles Duranton , Vernon Henderson J. , Strange William C. (eds) Handbook of Regional and Urban Economics SET , pp. 247 – 348 . Amsterdam : Elsevier . Combes P.-P. , Mayer T. , Thisse J.-F. ( 2008 ) Economic Geography: The Integration of Regions and Nations . Princeton, NJ : Princeton University Press . Crafts N. , Mulatu A. ( 2005 ) What explains the location of industry in Britain, 1871–1931? Journal of Economic Geography , 5 : 499 – 518 . Google Scholar CrossRef Search ADS Crafts N. , Mulatu A. ( 2006 ) How did the location of industry respond to falling transport costs in Britain before World War I? Journal of Economic History , 66 : 575 – 607 . Google Scholar CrossRef Search ADS Ellison G. , Glaeser E. L. ( 1999 ) The geographic concentration of industry: does natural advantage explain agglomeration? American Economic Review , 89 : 311 – 316 . Google Scholar CrossRef Search ADS Federico G. ( 2006 ) Industrial structure (1911–2001). In Renato Giannetti, , Michelangelo Vasta (eds) Evolution of Italian Enterprises in the 20th Century , pp. 15 – 47 . Heidelberg-New York : Springer . Google Scholar CrossRef Search ADS Federico G. , Natoli S. , Tattara G. , Vasta M. ( 2011 ) Il commercio estero italiano: 1862–1950, Collana storica della Banca d’Italia/Serie “Statistiche storiche”: Collana storica della Banca d’Italia, Laterza. Federico G. , Tena-Junguito A. ( 2014 ) The ripples of the industrial revolution: exports, economic growth, and regional integration in Italy in the early nineteenth century . European Review of Economic History , 18 : 349 – 369 . Google Scholar CrossRef Search ADS Felice E. ( 2013 ) Regional income inequality in Italy in the long run (1871–2001). Patterns and determinants. UHE Working Paper, Universitat Autonoma de Barcelona. Fenoaltea S. ( 2003 ) Peeking backward: regional aspects of industrial growth in post-unification Italy . Journal of Economic History , 63 : 1059 – 1102 . Fenoaltea S. ( 2004 ) Textile production in Italy’s regions, 1861–1913 . Rivista di Storia Economica , 20 : 145 – 174 . Fenoaltea S. ( 2011 ) The Reinterpretation of Italian Economic History: from Unification to the Great War . New York : Cambridge University Press . Google Scholar CrossRef Search ADS Fenoaltea S. ( 2016 ) The measurement of production: lessons from the engineering industry in Italy, 1911 . Research in Economic History , 32 : 73 – 145 . Google Scholar CrossRef Search ADS Fujita M. , Krugman P. , Venables A. ( 1999 ) The Spatial Economics: Cities, Regions and International Trade . Cambridge (MA) : MIT Press . Gelman A. , Hill J. ( 2006 ) Data Analysis using Regression and Multilevel/Hierarchical Models . Cambridge : Cambridge University Press . Google Scholar CrossRef Search ADS Harris C. D. ( 1954 ) The market as a factor in the localization of industry in the United States . Annals of the Association of American Geographers , 44 : 315 – 348 . Head K. , Mayer T. ( 2006 ) Regional wage and employment responses to market potential in the EU . Regional Science and Urban Economics , 36 : 573 – 594 . Google Scholar CrossRef Search ADS IGM . ( 2007 ) Classificazione dei Corsi d’acqua d’Italia. Florence: Istituto Geografico Militare. ISTAT . ( 1958 ) Sommario di Statistiche Storiche Italiane: 1861–1955 . Rome : ISTAT . Klein A. , Crafts N. ( 2012 ) Making sense of the manufacturing belt: determinants of US industrial location, 1880–1920 . Journal of Economic Geography , 12 : 775 – 807 . Google Scholar CrossRef Search ADS Krugman P. , Venables A. J. ( 1995 ) Globalization and the inequality of nations . The Quarterly Journal of Economics , 110 : 857 – 880 . Google Scholar CrossRef Search ADS Krugman P. R. ( 1991 ) Geography and Trade . Cambridge, MA : MIT Press . Lee D.-J. , Durbán M. ( 2011 ) P-spline ANOVA-type interaction models for spatio-temporal smoothing . Statistical Modelling , 11 : 49 – 69 . Google Scholar CrossRef Search ADS Martinez-Galarraga J. ( 2012 ) The determinants of industrial location in Spain, 1856–1929 . Explorations in Economic History , 49 : 255 – 275 . Google Scholar CrossRef Search ADS Martínez-Galarraga J. ( 2014 ) Market potential estimates in history: a survey of methods and an application to Spain, 1867–1930. EHES Working Papers in Economic History No. 51. Midelfart-Knarvik K. H. , Overman H. G. ( 2002 ) Delocation and European integration: is structural spending justified? Economic policy , 17 : 321 – 359 . Google Scholar CrossRef Search ADS Midelfart-Knarvik K. H. , Overman H. G. , Venables A. ( 2000 ) Comparative advantage and economic geography: estimating the determinants of industrial location in the EU . Centre for Economic Policy Research. Discussion Papers , No. 2618. Missiaia A. ( 2016 ) Where do we go from here? Market access and regional development in Italy (1871–1911) . European Review of Economic History , 20 : 215 – 241 . Google Scholar CrossRef Search ADS Puga D. ( 1999 ) The rise and fall of regional inequalities . European Economic Review , 43 : 303 – 334 . Google Scholar CrossRef Search ADS Redding S. , Venables A. J. ( 2004 ) Economic geography and international inequality . Journal of International Economics , 62 : 53 – 82 . Google Scholar CrossRef Search ADS Rosés J. R. ( 2003 ) Why isn’t the whole of Spain industrialized? New economic geography and early industrialization, 1797–1910 . Journal of Economic History , 63 : 995 – 1022 . Google Scholar CrossRef Search ADS Sundar V. J. , Ramesh R. , Rao P. S. , Saravanan P. , Sridharnath B. , Muralidharan C. ( 2001 ) Water management in leather industry . Journal of Scientific and Industrial Research , 60 : 443 – 450 . Tiefelsdorf M. , Griffith D. A. ( 2007 ) Semiparametric filtering of spatial autocorrelation: the eigenvector approach . Environment and Planning A , 39 : 1193 – 1221 . Google Scholar CrossRef Search ADS Tirado D. A. , Paluzie E. , Pons J. ( 2002 ) Economic integration and industrial location: the case of Spain before World War I . Journal of Economic Geography , 2 : 343 – 363 . Google Scholar CrossRef Search ADS Wolf N. ( 2007 ) Endowments vs. market potential: what explains the relocation of industry after the Polish reunification in 1918? Explorations in Economic History , 44 : 22 – 42 . Google Scholar CrossRef Search ADS Wood S. ( 2006 ) Generalized Additive Models: an Introduction with R . Boca Raton, Florida : CRC Press . Zamagni V. N. ( 1978 ) Industrializzazione e Squilibri Regionali in Italia: Bilancio dell’età Giolittiana . Bologna : Mulino . Appendix A.1. Map of Italy’s provinces at 1911 borders The name of each region (roughly NUTS-2 units), in bold, is followed by the name (and tag) of its provinces (roughly NUTS-3 units). Different shadows of gray simply highlight regional borders. There were 16 regions and 69 provinces. The list follows a broad North to South ordering. PIEDMONT: Alessandria (AL), Cuneo (CN), Novara (NO), Turin (TO) LIGURIA: Genoa (GE), Porto Maurizio (PM) LOMBARDY: Bergamo (BG), Brescia (BS), Como (CO), Cremona (CR), Mantua (MN), Milan (MI), Pavia (PV), Sondrio (SO) VENETIA: Belluno (BL), Padua (PD), Rovigo (RO), Treviso (TV), Udine (UD), Venice (VE), Verona (VR), Vicenza (VI) EMILIA: Bologna (BO), Ferrara (FE), Forlí (FO), Modena (MO), Parma (PR), Piacenza (PC), Ravenna(RA), Reggio Emilia (RE) TUSCANY: Arezzo (AR), Florence (FI), Grosseto (GR), Leghorn (LI), Lucca (LU), Massa Carrara (MS), Pisa (PI), Siena (SI) MARCHES: Ancona (AN), Ascoli Piceno (AP), Macerata (MC), Pesaro (PE) UMBRIA: Perugia (PG) LATIUM: Rome (RM) ABRUZZI: Aquila (AQ), Campobasso (CB), Chieti (CH), Teramo (TE) CAMPANIA: Avellino (AV), Benevento (BN), Caserta (CE), Naples (NA), Salerno (SA) APULIA: Bari (BA), Foggia (FG), Lecce (LE) BASILICATA: Potenza (PZ) CALABRIA: Catanzaro (CZ), Cosenza (CS), Reggio Calabria (RC) SICILY: Caltanissetta (CL), Catania (CT), Girgenti (AG), Messina (ME), Palermo (PA), Syracuse (SR), Trapani (TP) SARDINIA: Cagliari (CA), Sassari (SS) View largeDownload slide View largeDownload slide A.2. Data A.2.1. Value added for manufacturing sectors Provincial value added data at 1911 prices for the 12 manufacturing sectors listed in Table 1 are from Ciccarelli and Fenoaltea (2013). The data are available at http://onlinelibrary.wiley.com/doi/10/j.1468-0289.2011.00643.x/full (see in particular the Supplementary Materia). Ciccarelli and Fenoaltea obtained the provincial (i.e., NUTS 3) figures as follows. They first produced annual 1861–1913 regional (i.e., NUTS 2) value added data at 1911 prices as a result of a long-term project sponsored by the bank of Italy (Ciccarelli and Fenoaltea, 2009, 2014) (see https://www.bancaditalia.it/pubblicazioni/altre-pubblicazioni-storiche/produzione-industriale-1861-1913/). The regional estimates of industrial value added are based, whenever possible and depending on the availability of historical sources, on physical production data. To give an example, consider the chemical industry. The historical sources are extremely detailed and allow one to essentially track the 1861–1913 physical production of about 100 products (sulfuric acid, soda nitric acid, Leblanc hydrochloric acid, matches, superphosphate, Thomas slang, etc.) separately for each of the 16 regions. These physical production regional series (say, tons of output) are weighted by a unitarian value added (lire of value added per ton, or other physical unit) evaluated at the national level for the year 1911 and based on historical data on capital and wages. Regional 1861–1913 value added series at 1911 prices for the whole chemical sector are then obtained by summation of the separated regional series referring to the various products. The regional estimates thus account for differences in productivity among regions and industrial sectors. Once the regional value added data for each industrial sector (VAREGIO) were obtained, the authors allocated them to provinces (VAPROV), by using the provincial labor force shares (LFSHPROV) of regional totals, separately for each industrial sector. For each given region and time, provincial value added in province i and sector j has been obtained as: VAPROVi,j=VAREGIOj*LFSHPROVi,j. As a consequence, while annual 1861–1913 regional estimates (VAREGIO) are based on a rich set of detailed historical sources, provincial estimates (VAPROV) rest essentially on the information on the labor force as reported in the population censuses and, thus, are only available for the years 1871, 1881, 1901 and 1911 (the 1891 population census was not taken, and the first industrial census was carried out in 1911). Following this approach, we used the population censuses of 1871, 1881, 1901 and 1911 to further disaggregate at the provincial level the regional value added data for the textile and engineering industries. Detailed regional value added data for the textile industry and its sectoral components (wool, silk, cotton and other natural fibers) are reported in Fenoaltea (2004), while regional value added data for the engineering industry and its sectoral components (shipbuilding, machinery and blacksmith) are provided by Ciccarelli and Fenoaltea (2014). It is worth noticing that population censuses do not generally report information on industrial sectors, rather on individual professions. The latter, as is customary, must be mapped into industrial sectors, which always involves a certain degree of arbitrariness. The details are too numerous to be even partially illustrated here. However, just to give the basic idea, we allocated, province by province, and census year by census year, to the cotton sectors professions such as cotton spinner or cotton weaver, and to the blacksmith sector professions such as blacksmith or coppersmith. Finally, we illustrate the geographical distribution of location quotients. For the reason of space, we consider metalmaking and leather as representative of ‘high’ and ‘low’ K/L sectors, respectively. We consider instead separately the main sectoral components of the textile industry (cotton and silk). Figure A1 illustrates the case of ‘high’ K/L sectors in 1871 and 1911. There is a clear predominance of the West side on Central and Northern Italy. Among northern provinces, a marked difference between sub-Alpine provinces and those along the Po valley is evident. At the same time, however, some reallocation within the Center is also evident, with a shift away from the coastal Tyrrhenian regions (Latium and Southern Tuscany). In the South light colors prevail in 1871 and 1911, especially so along the backbone represented by the Central and Southern Apennines provinces down to Calabria, at the toe of Italy’s boot. Figure A1 View largeDownload slide Metalmaking (high K/L): choropleth maps of LQ values. (a) 1871 and (b) 1911. Figure A1 View largeDownload slide Metalmaking (high K/L): choropleth maps of LQ values. (a) 1871 and (b) 1911. Figure A2 considers ‘low’ K/L sectors in 1871 and 1911. There is little change here. In 1871, and even more in 1911, a clear North-South gradient or spatial trend is present. It is interesting to note that Figure A1 and Figure A2 essentially complement each other, suggesting possibly alternative driving forces for their industrial location. Figure A2 View largeDownload slide Leather (low K/L): choropleth maps of LQ values. (a) 1871 and (b) 1911. Figure A2 View largeDownload slide Leather (low K/L): choropleth maps of LQ values. (a) 1871 and (b) 1911. Figures A3 and A4 refer finally to the geographical distribution of cotton and silk, representing, together with metalmaking, the most agglomerated sectors in 1911. The predominant role of Alpine regions emerges clearly, with Piedmont and Lombardy alone accounting for about 75% of value added in textiles in 1913 (Fenoaltea, 2004). However, the sectoral reallocation between 1871 and 1911, with the whitening of the maps for the southern provinces of Campania and Sicily is also evident, especially for the cotton industry. These facts are largely consistent with the progressive mechanization of the cotton industry, also sustained by rising protection, with import duties increased during the late 1870s and 1880s (Fenoaltea, 2004). It is also interesting to note that silk producers were instead traditionally favorable to free trade (Cafagna, 1989; Fenoaltea, 2011). Figure A3 View largeDownload slide Cotton (high K/L): choropleth maps of LQ values. (a) 1871 and (b) 1911. Figure A3 View largeDownload slide Cotton (high K/L): choropleth maps of LQ values. (a) 1871 and (b) 1911. Figure A4 View largeDownload slide Silk (low K/L): choropleth maps of LQ values. (a) Year: 1871 and (b) Year: 1911. Figure A4 View largeDownload slide Silk (low K/L): choropleth maps of LQ values. (a) Year: 1871 and (b) Year: 1911. A.2.2. Gross domestic product Historical GDP estimates at the provincial level for the case of Italy are not available. Thus, we proxy for provincial GDP by allocating total regional GDP (NUTS-2 units) estimates for 1871, 1881, 1901 and 1911 to provinces (NUTS-3 units) using the provincial shares of regional population obtained by population censuses. Data on GDP at historical borders (NUTS-2 units) were kindly provided by E. Felice. GDP includes of course industry, agriculture and services. It is important to stress that the industrial component of the regional GDP estimates by Felice is largely based on the statistical reconstructions by Ciccarelli and Fenoaltea considered before [we refer the reader to Felice (2013) for further details]. A.3. Geoadditive model versus fixed effects model In this appendix, we compare the estimation results of the geoadditive model and the fixed effects model using as dependence variable the location quotient for the whole manufacturing sector. Starting from a model without explanatory variables, we report in Figure A5 the choropleth map of the estimated fixed effects, αî, from a simple model like ln⁡(LQi,t)=αi+εi,t along with the choropleth map of the predicted values of the spatial trend, sptî=f1̂+f2̂+f12̂,obtained from the estimation of a simple geoadditive model without explanatory variables, ln⁡(LQi,t)=α+f1(noi)+f2(ei)+f12(noi,ei)+εi,t. Figure A5 View largeDownload slide Whole manufacturing. Comparing estimated-fixed effects and spatial trend. (a) Estimated-fixed effects and (b) Estimated spatial trend. Figure A5 View largeDownload slide Whole manufacturing. Comparing estimated-fixed effects and spatial trend. (a) Estimated-fixed effects and (b) Estimated spatial trend. By comparing Figure A5 and Figure 2, we may conclude that both models capture well the spatial distribution of the location quotient for the whole manufacturing activity. Extending the two models to include the two time-varying variables (the log of Harris market potential and the log of labor skills), we get the results reported in Table A1 which confirm that the two approaches can be used as alternative specifications to control for unobserved spatial heterogeneity. The sign of the coefficients estimated with the two models is the same, although the magnitude and level of significance is a bit different. However, the model with spatial trend has an important advantage over the fixed effects model. In particular, the geoadditive model allows us to save degrees of freedom. The fixed effect model uses 69 + 2 = 71 degrees of freedom (one coefficient for each of the 69 regional dummies, plus two parameters for the two explanatory variables), while the model with the spatial trend uses 45 effective degrees of freedom (e.d.f. for the spatial trend plus three parameters for the intercept and the two explanatory variables). To estimate the geoadditive model, we used a number of knots equal to 14 for each of the two univariate components of the spatial trend—f1 and f2—and seven knots for the bivariate term—f12. Thus, in total we used 14 + 14 + 49 = 77 knots, that is, a large number of knots (larger than the number of fixed effects!) to better capture nonlinearities in the spatial trend. However, the penalized spline method reduced the number of effectively used parameters (i.e., the e.d.f.) to 45 (exactly 45.123). It is worth noticing that the degree of smoothing, that is the degree of penalization, is automatically determined by the REML estimation procedure, and thus there is no arbitrary choice from the researcher. This suggests that we do not need to remove all the between-group variability in the model to filter the spatial unobserved heterogeneity. What really matters is to remove any systematic spatial pattern from the residuals which might be correlated to the explanatory variables (i.e., the source of endogeneity). We conclude that the spatial trend model must be preferred to the fixed effects model as a way to control spatial heterogeneity and to get reasonable results of the effects of the explanatory variables. Table A1 Whole manufacturing Variable Fixed effects Geoadditive model Intercept −1.290*** (0.444) Ln(Mktpot) 0.096** 0.144*** (0.048) (0.047) Ln(Literacy) −0.262*** −0.363*** (0.054) (0.056) Variable Fixed effects Geoadditive model Intercept −1.290*** (0.444) Ln(Mktpot) 0.096** 0.144*** (0.048) (0.047) Ln(Literacy) −0.262*** −0.363*** (0.054) (0.056) Notes: Estimation results of the parametric-fixed effect model and of the semiparametric geoadditive model. Coefficients and standard errors (in parenthesis). Number of observations: 276 (69 provinces by four points in time). Table A1 Whole manufacturing Variable Fixed effects Geoadditive model Intercept −1.290*** (0.444) Ln(Mktpot) 0.096** 0.144*** (0.048) (0.047) Ln(Literacy) −0.262*** −0.363*** (0.054) (0.056) Variable Fixed effects Geoadditive model Intercept −1.290*** (0.444) Ln(Mktpot) 0.096** 0.144*** (0.048) (0.047) Ln(Literacy) −0.262*** −0.363*** (0.054) (0.056) Notes: Estimation results of the parametric-fixed effect model and of the semiparametric geoadditive model. Coefficients and standard errors (in parenthesis). Number of observations: 276 (69 provinces by four points in time). A.4. Specialization, concentration and spatial dependence A.4.1. Krugman specialization index We address the question of how specialized were Italian provinces by using Krugman’s specialization index Ki that, for a given point in time (t), is defined as: Ki=∑k|(sik−si−k)|0≤Ki≤2, where sik=(vik)/(∑kvik), and si−k is the share of industry k in the national total net of province i. Thus, the Krugman index provides a measure of the difference in the specialization of a given province when compared with the remaining provinces of the country. It takes the value of 0 if there is no difference, and the value of 2 if a province has no industries in common with the rest of Italy. Figure A6 illustrates the territorial distribution of Ki in 1871 and 1911. Figure A6 View largeDownload slide Specialization: Krugman index. Figure A6 View largeDownload slide Specialization: Krugman index. A.4.2. Theil index of concentration The relative Theil index of concentration (Ck) is based on the normalized location quotient, that is LQik normalized by the ratio between vi (manufacturing value added in province i) over V (manufacturing value added in Italy): Ck=∑iviVln⁡(LQik). The higher the value of Ck, the higher the concentration of industry k. The relative Theil index Ck provides useful information about the extent to which industries are concentrated in a limited number of areas, but it does not take into consideration whether those areas are close together or far apart. In other words, it does not take into account the spatial structure of the data. Every region is treated as an island, and its position in space relative to other regions is not taken into account. Thus, the relative Theil index, Ck(t), is an a-spatial measure of concentration: the same degree of concentration can be compatible with very different localization schemes. For example, two industries may appear equally geographically concentrated, while one is located in two neighboring regions, and the other splits between the northern and the southern part of the country. As pointed out by Arbia et al. (2013), a more accurate analysis of the spatial distribution of economic activities requires the combination of traditional measures of geographical concentration and methodologies that account for spatial dependence, in that they provide different and complementary information about the concentration of the various sectors. A.4.3. Moran’s I index of spatial dependence Spatial autocorrelation is present when the values of one variable observed at nearby locations are more similar than those observed in locations that are far apart. More precisely, positive spatial autocorrelation occurs when high or low values of a variable tend to cluster together in space and negative spatial autocorrelation when high values are surrounded by low values and vice versa. Among the spatial dependence measures the most widely used is the Moran’s I index based, as is well known, on a comparison of LQik at any location with the value of the same variable at surrounding locations. The most widely used is the Moran’s I index: I=(N∑i∑jwij)(∑i∑jwij(LQi−LQ¯)(LQj−LQ¯)∑i(LQi−LQ¯)2), (4) where N is the total number of provinces, LQi and LQj are the observed values of the location quotient for the locations i and j (with mean LQ¯), and the first term is a scaling constant. This statistic compares the value of a continuous variable at any location with the value of the same variable at surrounding locations. The spatial structure of the data is formally expressed in a spatial weight matrix W with generic elements wij (with i≠j). In this paper, we employ a row-standardized spatial weights matrix (W), whose elements wij on the main diagonal are set to zero whereas wij = 1 if dij<d¯ and wij = 0 if dij>d¯, with dij the great circle distance between the centroids of region i and region j and d¯ a cut-off distance (equal to 112 km, corresponding to the minimum distance which allows all provinces to have at least one neighbor). Table A2 reports the calculated value of the Theil and Moran indices for the years 1871, 1881, 1901 and 1911 with data disaggregated by sector. Table A2 Industry concentration: Theil index and Moran’s I (1) (2) (3) (4) (5) (6) (7) (8) Theil Moran 1871 1881 1901 1911 1871 1881 1901 1911 2.1 Foodstuffs 0.02 0.02 0.04 0.06 0.15 0.31 0.25 0.29 (−0.02) (0.00) (0.00) (0.00) 2.2 Tobacco 1.23 1.05 1.04 0.81 −0.04 0.00 0.02 0.01 (−0.63) (−0.41) (−0.32) (−0.38) 2.3 Textiles 0.26 0.30 0.45 0.45 0.33 0.38 0.47 0.49 (0.00) (0.00) (0.00) (0.00)     2.3.1 Cotton 0.90 0.91 0.70 0.63 0.04 0.08 0.39 0.44 (0.23) (0.11) (0.00) (0.00)     2.3.2 Wool 1.09 1.21 1.40 1.30 0.03 0.02 0.00 0.00 (0.27) (0.27) (0.42) (0.41)     2.3.3 Silk 1.44 1.45 1.14 1.04 0.04 0.04 0.13 0.26 (0.04) (0.08) (0.00) (0.00)     2.3.4 Other natural fibers 0.27 0.37 0.23 0.40 0.19 0.45 0.07 −0.06 (0.00) (0.00) (0.14) (0.73) 2.4 Clothing 0.11 0.13 0.10 0.11 0.29 0.30 0.40 0.35 (0.00) (0.00) (0.00) (0.00) 2.5 Wood 0.02 0.02 0.03 0.05 0.11 0.10 0.22 0.37 (−0.05) (−0.06) (0.00) (0.00) 2.6 Leather 0.05 0.06 0.11 0.16 0.60 0.62 0.69 0.73 (0.00) (0.00) (0.00) (0.00) 2.7 Metalmaking 0.38 0.57 0.86 0.74 0.28 0.26 0.17 0.17 (0.00) (0.00) (0.00) (0.00) 2.8 Engineering 0.05 0.03 0.06 0.07 0.17 0.19 0.12 0.02 (−0.01) (0.00) (−0.03) (−0.34)     2.8.1 Shipbuildings 1.95 1.56 1.67 1.79 0.00 −0.01 −0.02 −0.06 (0.34) (0.49) (0.55) (0.78)     2.8.2 Machinery 0.16 0.09 0.22 0.19 0.02 0.35 0.06 0.08 (0.32) (0.00) (0.16) (0.10)     2.8.3 Blacksmith 0.04 0.03 0.04 0.02 0.35 0.37 0.43 0.11 (0.00) (0.00) (0.00) (0.05) 2.9 Non-metallic mineral products 0.17 0.17 0.19 0.09 0.20 0.08 0.12 0.11 (0.00) (−0.03) (0.00) (−0.02) 2.10 Chemicals and rubber 0.18 0.16 0.23 0.17 −0.04 0.15 0.00 0.05 (−0.61) (−0.02) (−0.40) (−0.19) 2.11 Paper and printing 0.21 0.21 0.17 0.16 0.11 0.07 0.06 0.07 (−0.05) (−0.12) (−0.17) (−0.13) 2.12 Sundry 0.33 0.89 0.62 0.46 0.05 0.00 0.01 0.15 (−0.19) (−0.37) (−0.36) (−0.01) 2. Total manufacturing 0.03 0.04 0.07 0.08 0.05 0.00 0.01 0.15 (−0.47) (−0.41) (−0.09) (−0.02) (1) (2) (3) (4) (5) (6) (7) (8) Theil Moran 1871 1881 1901 1911 1871 1881 1901 1911 2.1 Foodstuffs 0.02 0.02 0.04 0.06 0.15 0.31 0.25 0.29 (−0.02) (0.00) (0.00) (0.00) 2.2 Tobacco 1.23 1.05 1.04 0.81 −0.04 0.00 0.02 0.01 (−0.63) (−0.41) (−0.32) (−0.38) 2.3 Textiles 0.26 0.30 0.45 0.45 0.33 0.38 0.47 0.49 (0.00) (0.00) (0.00) (0.00)     2.3.1 Cotton 0.90 0.91 0.70 0.63 0.04 0.08 0.39 0.44 (0.23) (0.11) (0.00) (0.00)     2.3.2 Wool 1.09 1.21 1.40 1.30 0.03 0.02 0.00 0.00 (0.27) (0.27) (0.42) (0.41)     2.3.3 Silk 1.44 1.45 1.14 1.04 0.04 0.04 0.13 0.26 (0.04) (0.08) (0.00) (0.00)     2.3.4 Other natural fibers 0.27 0.37 0.23 0.40 0.19 0.45 0.07 −0.06 (0.00) (0.00) (0.14) (0.73) 2.4 Clothing 0.11 0.13 0.10 0.11 0.29 0.30 0.40 0.35 (0.00) (0.00) (0.00) (0.00) 2.5 Wood 0.02 0.02 0.03 0.05 0.11 0.10 0.22 0.37 (−0.05) (−0.06) (0.00) (0.00) 2.6 Leather 0.05 0.06 0.11 0.16 0.60 0.62 0.69 0.73 (0.00) (0.00) (0.00) (0.00) 2.7 Metalmaking 0.38 0.57 0.86 0.74 0.28 0.26 0.17 0.17 (0.00) (0.00) (0.00) (0.00) 2.8 Engineering 0.05 0.03 0.06 0.07 0.17 0.19 0.12 0.02 (−0.01) (0.00) (−0.03) (−0.34)     2.8.1 Shipbuildings 1.95 1.56 1.67 1.79 0.00 −0.01 −0.02 −0.06 (0.34) (0.49) (0.55) (0.78)     2.8.2 Machinery 0.16 0.09 0.22 0.19 0.02 0.35 0.06 0.08 (0.32) (0.00) (0.16) (0.10)     2.8.3 Blacksmith 0.04 0.03 0.04 0.02 0.35 0.37 0.43 0.11 (0.00) (0.00) (0.00) (0.05) 2.9 Non-metallic mineral products 0.17 0.17 0.19 0.09 0.20 0.08 0.12 0.11 (0.00) (−0.03) (0.00) (−0.02) 2.10 Chemicals and rubber 0.18 0.16 0.23 0.17 −0.04 0.15 0.00 0.05 (−0.61) (−0.02) (−0.40) (−0.19) 2.11 Paper and printing 0.21 0.21 0.17 0.16 0.11 0.07 0.06 0.07 (−0.05) (−0.12) (−0.17) (−0.13) 2.12 Sundry 0.33 0.89 0.62 0.46 0.05 0.00 0.01 0.15 (−0.19) (−0.37) (−0.36) (−0.01) 2. Total manufacturing 0.03 0.04 0.07 0.08 0.05 0.00 0.01 0.15 (−0.47) (−0.41) (−0.09) (−0.02) Note: p-values in parenthesis. Table A2 Industry concentration: Theil index and Moran’s I (1) (2) (3) (4) (5) (6) (7) (8) Theil Moran 1871 1881 1901 1911 1871 1881 1901 1911 2.1 Foodstuffs 0.02 0.02 0.04 0.06 0.15 0.31 0.25 0.29 (−0.02) (0.00) (0.00) (0.00) 2.2 Tobacco 1.23 1.05 1.04 0.81 −0.04 0.00 0.02 0.01 (−0.63) (−0.41) (−0.32) (−0.38) 2.3 Textiles 0.26 0.30 0.45 0.45 0.33 0.38 0.47 0.49 (0.00) (0.00) (0.00) (0.00)     2.3.1 Cotton 0.90 0.91 0.70 0.63 0.04 0.08 0.39 0.44 (0.23) (0.11) (0.00) (0.00)     2.3.2 Wool 1.09 1.21 1.40 1.30 0.03 0.02 0.00 0.00 (0.27) (0.27) (0.42) (0.41)     2.3.3 Silk 1.44 1.45 1.14 1.04 0.04 0.04 0.13 0.26 (0.04) (0.08) (0.00) (0.00)     2.3.4 Other natural fibers 0.27 0.37 0.23 0.40 0.19 0.45 0.07 −0.06 (0.00) (0.00) (0.14) (0.73) 2.4 Clothing 0.11 0.13 0.10 0.11 0.29 0.30 0.40 0.35 (0.00) (0.00) (0.00) (0.00) 2.5 Wood 0.02 0.02 0.03 0.05 0.11 0.10 0.22 0.37 (−0.05) (−0.06) (0.00) (0.00) 2.6 Leather 0.05 0.06 0.11 0.16 0.60 0.62 0.69 0.73 (0.00) (0.00) (0.00) (0.00) 2.7 Metalmaking 0.38 0.57 0.86 0.74 0.28 0.26 0.17 0.17 (0.00) (0.00) (0.00) (0.00) 2.8 Engineering 0.05 0.03 0.06 0.07 0.17 0.19 0.12 0.02 (−0.01) (0.00) (−0.03) (−0.34)     2.8.1 Shipbuildings 1.95 1.56 1.67 1.79 0.00 −0.01 −0.02 −0.06 (0.34) (0.49) (0.55) (0.78)     2.8.2 Machinery 0.16 0.09 0.22 0.19 0.02 0.35 0.06 0.08 (0.32) (0.00) (0.16) (0.10)     2.8.3 Blacksmith 0.04 0.03 0.04 0.02 0.35 0.37 0.43 0.11 (0.00) (0.00) (0.00) (0.05) 2.9 Non-metallic mineral products 0.17 0.17 0.19 0.09 0.20 0.08 0.12 0.11 (0.00) (−0.03) (0.00) (−0.02) 2.10 Chemicals and rubber 0.18 0.16 0.23 0.17 −0.04 0.15 0.00 0.05 (−0.61) (−0.02) (−0.40) (−0.19) 2.11 Paper and printing 0.21 0.21 0.17 0.16 0.11 0.07 0.06 0.07 (−0.05) (−0.12) (−0.17) (−0.13) 2.12 Sundry 0.33 0.89 0.62 0.46 0.05 0.00 0.01 0.15 (−0.19) (−0.37) (−0.36) (−0.01) 2. Total manufacturing 0.03 0.04 0.07 0.08 0.05 0.00 0.01 0.15 (−0.47) (−0.41) (−0.09) (−0.02) (1) (2) (3) (4) (5) (6) (7) (8) Theil Moran 1871 1881 1901 1911 1871 1881 1901 1911 2.1 Foodstuffs 0.02 0.02 0.04 0.06 0.15 0.31 0.25 0.29 (−0.02) (0.00) (0.00) (0.00) 2.2 Tobacco 1.23 1.05 1.04 0.81 −0.04 0.00 0.02 0.01 (−0.63) (−0.41) (−0.32) (−0.38) 2.3 Textiles 0.26 0.30 0.45 0.45 0.33 0.38 0.47 0.49 (0.00) (0.00) (0.00) (0.00)     2.3.1 Cotton 0.90 0.91 0.70 0.63 0.04 0.08 0.39 0.44 (0.23) (0.11) (0.00) (0.00)     2.3.2 Wool 1.09 1.21 1.40 1.30 0.03 0.02 0.00 0.00 (0.27) (0.27) (0.42) (0.41)     2.3.3 Silk 1.44 1.45 1.14 1.04 0.04 0.04 0.13 0.26 (0.04) (0.08) (0.00) (0.00)     2.3.4 Other natural fibers 0.27 0.37 0.23 0.40 0.19 0.45 0.07 −0.06 (0.00) (0.00) (0.14) (0.73) 2.4 Clothing 0.11 0.13 0.10 0.11 0.29 0.30 0.40 0.35 (0.00) (0.00) (0.00) (0.00) 2.5 Wood 0.02 0.02 0.03 0.05 0.11 0.10 0.22 0.37 (−0.05) (−0.06) (0.00) (0.00) 2.6 Leather 0.05 0.06 0.11 0.16 0.60 0.62 0.69 0.73 (0.00) (0.00) (0.00) (0.00) 2.7 Metalmaking 0.38 0.57 0.86 0.74 0.28 0.26 0.17 0.17 (0.00) (0.00) (0.00) (0.00) 2.8 Engineering 0.05 0.03 0.06 0.07 0.17 0.19 0.12 0.02 (−0.01) (0.00) (−0.03) (−0.34)     2.8.1 Shipbuildings 1.95 1.56 1.67 1.79 0.00 −0.01 −0.02 −0.06 (0.34) (0.49) (0.55) (0.78)     2.8.2 Machinery 0.16 0.09 0.22 0.19 0.02 0.35 0.06 0.08 (0.32) (0.00) (0.16) (0.10)     2.8.3 Blacksmith 0.04 0.03 0.04 0.02 0.35 0.37 0.43 0.11 (0.00) (0.00) (0.00) (0.05) 2.9 Non-metallic mineral products 0.17 0.17 0.19 0.09 0.20 0.08 0.12 0.11 (0.00) (−0.03) (0.00) (−0.02) 2.10 Chemicals and rubber 0.18 0.16 0.23 0.17 −0.04 0.15 0.00 0.05 (−0.61) (−0.02) (−0.40) (−0.19) 2.11 Paper and printing 0.21 0.21 0.17 0.16 0.11 0.07 0.06 0.07 (−0.05) (−0.12) (−0.17) (−0.13) 2.12 Sundry 0.33 0.89 0.62 0.46 0.05 0.00 0.01 0.15 (−0.19) (−0.37) (−0.36) (−0.01) 2. Total manufacturing 0.03 0.04 0.07 0.08 0.05 0.00 0.01 0.15 (−0.47) (−0.41) (−0.09) (−0.02) Note: p-values in parenthesis. A.5. The sectorial effects of Alpine regions as a function of river The following figure shows the estimated marginal effects of the dummy Alpine, conditional on the effect of market potential and literacy, as a function of the river variable. Alpine provinces show a comparative advantage in the location of industrial activity in the case of machinery, cotton, silk, blacksmith and non-metallic mineral products. Figure A7 View largeDownload slide Estimated coefficient of Alpine by ln(River) with simulated 95% confidence intervals. Figure A7 View largeDownload slide Estimated coefficient of Alpine by ln(River) with simulated 95% confidence intervals. © The Author (2017). Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

Journal of Economic Geography – Oxford University Press

Published: Oct 9, 2017

There are no references for this article.