Abstract

This paper sheds new light on the agricultural side of the Italian regional divide from an economic geography perspective, following a Von Thünen approach. The central hypothesis is that the development of the nonagricultural economy in the Northern cities drove the location of agricultural output and inputs during the interwar years. A new database on Italian agriculture around 1930 fully confirms the key role of access to domestic markets in shaping agricultural activity. Thus, the causes of Southern agriculture falling behind are revealed: it is not very surprising that an agricultural divergence joined an already ongoing industrial divergence during a period in which international markets collapsed. It was the growth of Northern industry that led to the growth of Northern agriculture, and not vice versa.

Introduction

This paper takes an economic geography approach to explore the patterns of Italian interwar agriculture and its role in the long-lasting Italian regional divide. After framing the Southern falling behind, I turn to an explanation of the divergence exclusively based on factor accumulation, building upon a neoclassical version of the Von Thünen model of land use. Hence, I introduce a new cross sectional database of Italian agriculture around 1930 at a fine geographical level of disaggregation. Using this database, I first estimate the agricultural total factor productivities (TFP) throughout the country, which suggest that efficiency differences did not drive the Southern agricultural failure during the interwar years. Indeed, econometric results indicate that the geographical allocation of factors of production was shaped by the access to domestic nonagricultural markets, which in turn shaped output and land rents, as predicted by the Von Thünen model. Hence, the initial industrial divergence is causally connected with the subsequent agricultural one by a simple demand mechanism: it was the growth of Northern industry that led to the growth of Northern agriculture, and not vice versa.

The Italian agricultural divergence

Southern Italy, the so-called Mezzogiorno, is often cited as a paradigm of convergence failure. In the European context, it may be “the” paradigm of convergence failure. As such, it has been object of scholarly debates for decades. However, systematic quantitative research on the North–South divergence has only been produced in recent years (Felice 2011). The new picture emerging from the historical reconstruction of national accounts at regional level roughly confirms the conventional wisdom of a Southern economic failure, but the timing of the process and its components are substantially reshaped. Thanks to an initially successful agricultural performance, the income of Southern Italians was only slightly below the national average as late as the end of the nineteenth century. The emergence of an income gap between the two main sections of the country is now considered much more of an issue of the interwar years than a direct consequence of Unification in 1861 (Brunetti et al. 2011; Daniele and Malanima 2011; Iuzzolino et al. 2011; Vecchi 2011).

Even more challenging to the conventional wisdom is the breakdown by sector of the Italian divergence. Agriculture was traditionally considered the villain of the story: a poor Southern agricultural performance, usually attributed to inefficient institutions (Sereni 1977), kept incomes low, was unable to release employment to the rest of the economy and, ultimately, inhibited the industrial development of the Southern economy (Zamagni 1993). This causality now seems implausible (Cohen and Federico 2001). Agricultural labor productivity was actually higher in the South than in the North towards the turn of the century (Federico 2003a, b), when an industrial gap had already emerged.

It was only during the interwar years that Southern agriculture experienced a growth failure: output increasingly concentrated in Northern regions (see figure 1), while employment shares remained constant throughout the country. Indeed, while Northern regions industrialized, during the first half of the twentieth-century the Southern regions not only failed to do so but even lost their initial relative leadership in the agricultural sector (see figure 2). This late and unexplained agricultural falling behind joined the industrial divergence and caused the extreme regional polarization in income per capita between the Northern and Southern regions, a polarization from which the country, after a limited postwar convergence led by massive state intervention, has never recovered.

Figure 1.

Southern Italy's agriculture falling behind (I)—declining share of national output. Source: Own elaboration based on regional output estimates, at constant 1911 prices, by Federico (2003a, b). In order to assure intertemporal comparability, National output is computed excluding in 1938 and 1951 the Northern acquisitions after World War I, i.e., the regions of Venezia Tridentina/Trentino-Alto Adige and Venezia Giulia/Friuli-Venezia Giulia (the latter partially lost after World War II). The province of Udine, accounting for approximately 1.5% of national output in 1936–1938, was transferred from Venetia to FVG after World War II, and hence the share of the Northern regions in National output in 1951 is actually slightly underestimated.

Figure 1.

Southern Italy's agriculture falling behind (I)—declining share of national output. Source: Own elaboration based on regional output estimates, at constant 1911 prices, by Federico (2003a, b). In order to assure intertemporal comparability, National output is computed excluding in 1938 and 1951 the Northern acquisitions after World War I, i.e., the regions of Venezia Tridentina/Trentino-Alto Adige and Venezia Giulia/Friuli-Venezia Giulia (the latter partially lost after World War II). The province of Udine, accounting for approximately 1.5% of national output in 1936–1938, was transferred from Venetia to FVG after World War II, and hence the share of the Northern regions in National output in 1951 is actually slightly underestimated.

Figure 2.

Southern Italy's agriculture falling behind (II)—declining relative labor productivity (Italy = 1). Source: Own elaboration based on Federico (2003a, b) for output (GSP) and Vitali (1968) for labor force. National and Northern Labor Productivity are computed excluding in 1938 and 1951 the Northern acquisitions after World War I.

Figure 2.

Southern Italy's agriculture falling behind (II)—declining relative labor productivity (Italy = 1). Source: Own elaboration based on Federico (2003a, b) for output (GSP) and Vitali (1968) for labor force. National and Northern Labor Productivity are computed excluding in 1938 and 1951 the Northern acquisitions after World War I.

Thus, the reason why this agricultural falling behind occurred is now the real big question of Italian postunitary agrarian history. Every divergence in growth rates necessarily has as proximate causes either (i) a growing technological gap, (ii) a differential in the rates of factor accumulation, or (iii) a combination of the previous two. For the Italian case, the contribution of either factor is unknown.1 In any case, whatever the relative importance of factor accumulation and TFP in explaining the Southern agricultural divergence, it is worth bearing in mind that these were just proximate causes: they tell us how Northern and Southern agricultures diverged, not why. The ultimate causes are even more mysterious, although they are the really important ones. Why should either factor accumulation or technical progress diverge so systematically for such a long period across the main sections of the country?

In this paper I explore a simple mechanism for the falling behind of Southern agriculture. As an ultimate cause, I focus on the access to domestic markets for agricultural products during a period of distress in international trade. In a Von Thünen model of land use market access acts as the ultimate cause that drives factor accumulation as a proximate cause of the divergence. Despite it being unnecessary to think that the process was driven by a single cause, access to markets is a good candidate for having been the main actor. It directly links the two sides of the Italian regional divide, the rise of industry in the North and the agricultural divergence, into a single story. As Northern cities and their hinterlands industrialized, the number of people employed in nonagricultural sectors grew. These people had to be fed. Farmers in areas closer to the sources of a rising demand for their products were able and willing to further expand output, and they did so with a higher input use. Factor accumulation allowed the agricultural supply to expand to meet the growing agricultural demand. As the mechanism seems to fit well with the time and sector profile of the process, its explicative power is especially attractive and it is worth exploring.

Because we can rely on well-established economic theories about the relationship between access to markets and the spatial patterns of agricultural activity, a formal model can be used to derive testable predictions. Because lack of data makes it impossible to directly test a dynamic version of the model at the present stage of the research, it is necessary to pursue a different research strategy. By testing several predictions derived from the theoretical model in its static version, it is possible to verify whether it is an adequate description of the Italian agriculture at the midpoint of the divergence or not, provided a database exists that is sufficiently large and precise. In Section 3, I describe the model, the main predictions and their relationship with the available data. Later, I provide formal statistical tests of the main testable hypotheses and, after including some robustness checks, I discuss them. Finally, I conclude.

Analytical framework: access to markets and spatial patterns of agricultural activity

Economic geography models are usually used to explain the uneven spatial distribution of economic activity. Their use in addressing historical questions is relatively recent. Nonetheless, the increasing interest of economic historians in economic geography has been primarily focused on the manufacturing sector (Kim 1995; Wolf 2007; Klein and Crafts 2012; Martínez-Galarraga 2012). Because these types of model rely mainly on the seminal framework laid out by Krugman (1991) and further refinements of it (as Puga 1999), i.e., the so-called New Economic Geography (NEG), their focus is basically on sectors potentially having increasing returns (as industry). Agriculture is occasionally involved in the empirical analysis, but it is merely as an exogenous determinant of income, with industry generally being the main subject of interest (Rosés et al. 2010; Combes et al. 2011).

The Italian case, perhaps the paradigm of spatial polarization of economic activity in Europe, constitutes a somewhat surprising absence in this brand of literature. The first attempt to explicitly adopt a NEG framework in order to explain the long-run performance of the Italian regions is A'Hearn and Venables (2011).2 Their explorative paper states that access to markets was a key determinant of the growth pattern of the Italian regions. According to their framework, the North was better suited than the South for taking advantage of the specific conditions of every successive period since unity. If natural advantage was the main determinant of economic activity from 1861 to 1890, during 1890–1950 the key was access to domestic markets and after 1950 it was access to international markets. Nonetheless, their main focus is, again, on industry, and the importance of industry in explaining the North–South divergence. This leaves both the increasing concentration of agricultural output in the Northern regions and the productivity divergence unexplained.

This lack of interest of the economic geography models in the agricultural sector seems somewhat paradoxical when we take into account that the first model of spatial distribution of economic activity precisely focused on agriculture: the Von Thünen model (Von Thünen 1826) aimed at explaining the patterns of land use around a central city. Writing in the first half of the nineteenth century, Von Thünen's work was rather neglected for decades, at least outside the specific field of urban economics (Krugman 1991). In the second half of the twentieth century the Von Thünen model has been refined with a mathematically rigorous formulation within the neoclassical framework (Beckmann 1972; Samuelson 1983), but its empirical applications have still been scarce. Specifically, the model has not found much application in historical economics. The recent contribution of Kopsidis and Wolf (2012) makes use of a Thünen-inspired framework in a historical economic analysis for the first time. In their pioneering paper access to urban markets is found to have been a key determinant of agricultural patterns across nineteenth-century Prussia, explaining the East–West gradient in Prussian land productivity.

In this paper I address the question of the interwar agricultural divergence of the Italian regions strongly relying on Kopsidis and Wolf (2012) and in turn partly inspired by Beckmann (1972). I have slightly modified the model in order to better adapt it to the problem at hand. In particular, (i) because the composition of output is not the main concern, I impose without loss of generality a single-product production function and (ii) I include agricultural capital as an input. The main components of the model are as follows.

First, a central market city with a given population N is exogenously located in a featureless plain. For the sake of simplicity agricultural production is represented by a single commodity Q with a unitary price at the central market p.

In any location situated at a distance d from the central market, agricultural output is produced with a Cobb–Douglas production function:  

(1)
$$Q = A{K^\alpha }{L^\beta }{T^\gamma },\quad (\alpha + \beta + \gamma = 1)$$

Agricultural production uses capital K, labor L, and land T. A is a scalar capturing the exogenous differences in efficiency and environment; one might think of it as a composite variable capturing the effects on the productive factors of the human-origin technology as well as of physical and environmental conditions. Because any unit (for example, hectares) of land represents a location and is fixed, it will be useful to represent the production function in per hectare terms by dividing both sides of (1) by T:  

(2)
$$q = A{k^\alpha }{l^\beta }$$

The price of agricultural output at the central market is an increasing function of the city's population:  

(3)
$$p = f\,(N ),\quad f^{\prime} \gt 0$$

At every location situated at a distance d from the central market, the farm-gate prices pd must discount transport costs. Transport costs are assumed to be of a general form, with an ad valorem (t1) component and a per unit (t2) component:  

(4)
$$p = {\,p_d}({1 + {t_1}d} )+ {t_2}d \quad \hbox{or}\quad{\,p_d} = \displaystyle{{\,p - {t_2}d} \over {(1 + {t_1}d)}} = \displaystyle{{\,f\,(N) - {t_2}d} \over {(1 + {t_1}d)}}$$

The farm-gate prices prices at every location are increasing in the population and decreasing in distance from the central market city.

At every location, rent per hectare is defined as  

(5)
$${R_d} = {\,p_d}q - rk - wl = y - rk - wl$$
that is, output valued at farm-gate prices (y) minus production costs.

The landowner maximization of Rd yields the landowner factor demands which, given factors prices, denote the equilibrium factor intensity:  

(6)
$${l^{*}_{d}} = {\left( {\displaystyle{\alpha \over r}} \right)^{\alpha /\gamma }}{\left( {\displaystyle{\beta \over w}} \right)^{(1 - \alpha) /\gamma }}{A^{1/\gamma }}{\,p_d}^{1/\gamma } $$
 
(7)
$${k^{*}_{d}} = {\left( {\displaystyle{\alpha \over r}} \right)^{(1 - \beta) /\gamma }}{\left( {\displaystyle{\beta \over w}} \right)^{\beta /\gamma }}{A^{1/\gamma }}{\,p_d}^{1/\gamma } $$
where the subscript d emphasizes that factor intensity at every location d is a function of its distance from the market city. Similarly, physical output per hectare at every location is  
(8)
$${q^{*}_{d}} = A{k^*}^\alpha {l^*}^\beta = {\left( {\displaystyle{\alpha \over r}} \right)^{\alpha /\gamma }}{\left( {\displaystyle{\beta \over w}} \right)^{\beta /\gamma }}{A^{1/\gamma }}{\,p_d}^{(1 - \gamma) /\gamma } $$
And the value of output per hectare is  
(9)
$$\; {y^{*}_{d}} = {\,p_d}{q_d}^* = {\left( {\displaystyle{\alpha \over r}} \right)^{\alpha /\gamma }}{\left( {\displaystyle{\beta \over w}} \right)^{\beta /\gamma }}{A^{1/\gamma }}{\,p_d}^{1/\gamma } $$

It follows that equilibrium rent per hectare is  

(10)
$${R^{*}_{d}} = \; {y^{*}_{d}} - \; r{k^{*}_{d}} - w{l^{*}_{d}} = \gamma {\left( {\displaystyle{\alpha \over r}} \right)^{\alpha /\gamma }}{\left( {\displaystyle{\beta \over w}} \right)^{\beta /\gamma }}{A^{1/\gamma }}{\,p_d}^{1/\gamma } $$

Four key variables of the model (i.e., labor per hectare, capital per hectare, the value of output per hectare and rents per hectare) are all in equilibrium increasing functions of the central city size. At the same time, they are all decreasing functions of distance to the city. This means that agricultural activity will be spatially distributed following a gradient of decreasing intensity around the central city. Rents, mobile inputs and output per hectare will all follow such a pattern. Moreover, physical output per hectare will be higher the closer we get to the city. The last fact is important from an empirical point of view: even if output is valued imperfectly due to the unavailability of farm-gate prices, it will follow an intensity gradient around the city.

Equations (6), (7), (9), and (10) provide a straightforward strategy to test the Von Thünen hypothesis. If access to markets was driving agricultural production, we must find all of the four aforementioned variables to be increasing in market city size and decreasing in distance to it.

Data: a new database on Italian agriculture circa 1930: output, rents, labor, capital, and TFP

The analysis outlined in Section 3 requires data on the four key variables of the model: output, capital, labor, and rents. As has been said, no estimates of agricultural output or capital exist at a level of disaggregation lower than region. Relying on an unexploited set of sources, I have produced the first ever estimates of Italian historical agricultural output and capital at a level of disaggregation lower than region, namely at the agrarian zone level (a statistical unit used during the interwar years). In this section I briefly introduce the dataset. Further details on the estimation procedures can be found in the Supplementary material, Appendix.

The agrarian zone is the unit of observation suggested by Federico (2007) as optimal in dealing with Italian agriculture, although he deemed the task of collecting data “practically impossible” at such a level to carry out statistical analysis. Indeed, data at the agrarian zone level were not regularly published by the main Italian statistical services. There is a single point in time for which detailed statistical information is available at such a level of disaggregation: the years around 1930. Fortunately for the purposes of this paper, 1930 is a sort of middle-point for the period under scrutiny, when the falling behind of Southern agriculture was well underway.

The estimate of the agricultural output comprises sixty-eight products which constituted in 1938 close to 85 percent of all the agricultural output (according to the national estimate of Federico 2000). The data on agricultural output can be estimated from a couple of official inquiries: the Agrarian Cadastre of 1929 (ISTAT 1932–1937) and the Livestock Census of 1930 (ISTAT 1933–1935). The Agrarian Cadastre was a massive inquiry into Italian agriculture carried out in 1929 and published in more than 11,000 pages. It reported the area and yield of every single crop in every single municipality of Italy as they were in 1929. The Livestock Census reflected the stock of every animal, with breakdown by age, sex, and function, in March 1930. I assume that the animal stock did not change significantly between 1929 and March 1930. From these two sources I estimate the gross saleable production (GSP) at the agrarian zone level in 1929 by and large following the procedures noted in Federico (1992, 2000).

Basically, the estimation procedure consists of: (i) subtracting re-uses and seeds from the gross output of vegetable products using coefficients found in several different sources, generally published between the 1920s and 1940s (official statistics, agricultural handbooks and actual farm surveys); (ii) transforming the initial physical output into the final product by using coefficients deduced from the closer available data (usually the late 1930s) when the product involved some transformation, as in the cases of wine and oil; (iii) estimating the physical output from the pertinent stock in the case of several animal products, using coefficients selected with criteria and from sources similar to those employed in the estimation of other products; (iv) valuing the physical output at the average market prices during the harvest period, which are assumed to be the closest observable measure of the unobservable farm-gate prices. As a general rule, I have used as many different prices and coefficients as made possible by source availability in terms of geographical disaggregation, as well as in terms of the age, race, and sex structure of the animal stock. Output is valued at 1938 prices, because data on some products are missing for previous years and also because the choice of 1938 allows comparability with land rents (see below).

For the capital estimate some items are available at the agrarian zone level from a variety of sources (including the aforementioned ones): livestock (twenty-one varieties), trees (nineteen species), irrigation, land reclamation, and terracing capital. Nonresidential buildings and structures (such as stables, barns, hay lofts, manure depots, wine and oil facilities, etc.) are not available and are thus estimated by the demand side, again using coefficients from technical handbooks or farm surveys. Machinery, fertilizers, pesticides and fuel consumption are available from the official statistics at provincial level and have been assigned at the agrarian zone level according to informed criteria. The capital estimate is at 1938 prices, like the output one.

The resulting estimates, in per hectare terms, can be appreciated in Map 1 and Map 2. This new database, with 793 units of observation as against the preexisting estimates for just eighteen regions, enhances our understanding of Italian prewar agriculture to an extent usually hidden by the regional level of aggregation. While output and capital per hectare were generally higher in the North than in the South, there were areas outside the Po-Valley with significantly high values of both variables and the Northern plains themselves reveal some interesting patterns. First, high output per hectare was found (perhaps without much surprise) in all the Po- Valley, with peaks of land productivity and capital intensity to be found in its Western half (in an area roughly centered in the vicinity of Milan) and along the Via Emilia. Second, the Arno Valley in Tuscany and the coasts of Liguria were the other areas in the Center-North outside the Po-Valley that exhibited high levels of output and capital per hectare. Third, in the South a noticeably intensive agriculture was located around its biggest cities and ports: in the area centered around Naples, in the coastal strip from Brindisi to Bari in Apulia, in the so-called Conca d'Oro around Palermo, in the area centered around Messina and Catania (in the Eastern coast of the Sicily) and Reggio Calabria in the continent. Fourth, a key difference between the Northern and the Southern areas of intensive agriculture is that the former spread well beyond the immediate hinterland of the largest urban agglomerations.

Map 1.

Agricultural output per hectare in 1929–1930. Output measured as Gross Saleable Production valued at 1938 prices.

Map 1.

Agricultural output per hectare in 1929–1930. Output measured as Gross Saleable Production valued at 1938 prices.

Map 2.

Agricultural capital per hectare in 1929–1930. Capital stock valued at 1938 prices.

Map 2.

Agricultural capital per hectare in 1929–1930. Capital stock valued at 1938 prices.

The other variables required by the analysis (labor and land rents) are available from ready-made sources. For the agricultural labor, I rely on the Population Census taken in 1931 (ISTAT 1931–1935). While employment in agriculture was not published at the agrarian zone level in the Census, both the number of families whose family head was employed in agriculture and the number of members of such families were. It may be observed that there are certain concerns in the literature about the underestimation of female agricultural employment in Italian Census (Vitali 1968). For the problem at hand, the use of the two available measures of agricultural labor can be considered a robustness check and a way to overcome these measurement problems, because both measures represent the lower (family heads) and the upper (family members) bounds of employment.

For rents, I rely on the first simultaneous estimate of land rents made with homogeneous criteria3 and published in an official inquiry on landownership (INEA 1946–1948). The estimate was made for fiscal purposes and was aimed at computing the economic rent, i.e., the actual return to land after deducting all imputable variable and fixed costs, including the imputed cost of the owner's work. Rents were estimated at the 1937–39 local prices reflecting the productive conditions of those years.

With this database it is possible to tentatively turn from partial to total factor productivity, which is a crucial issue in the North–South agricultural divergence. If the production function is of a Cobb–Douglas form, then the TFP of the i-th agrarian zone is given by: 

$$\hbox{TF}{\hbox{P}_i} = {A_i} = \displaystyle{{{Y_i}} \over {\mathop \prod \nolimits_{\,j = 1}^n {K_i}{{_j }^{{\alpha _j}}}{L_i}^\beta {T_i}^\gamma }}\; $$

It is possible to use the data described previously to obtain estimates of the coefficients αj (for all the n forms of agricultural capital, with j = 1,…, n), β, and γ, which in turn allow to obtain a measure of TFP for every single agrarian zone. The ordinary least squares (OLS) estimates of the production function of Italian agriculture around 1930 are presented in table 1.

Table 1.

The production function of Italian agriculture (1930 ca.)

Dep. var.: agricultural output OLS (1) OLS (2) OLS (3) OLS (4) 
Capital 0.775*** (0.026) 0.733*** (0.029)   
Labor (families) 0.232*** (0.035)  0.446*** (0.035)  
Labor (individuals)  0.264*** (0.034)  0.420*** (0.035) 
Land 0.049** (0.023) 0.056*** (0.022) 0.111*** (0.030) 0.162*** (0.030) 
Reclamation   0.005*** (0.002) 0.005*** (0.002) 
Irrigation   0.009*** (0.002) 0.010*** (0.002) 
Working capital   0.079*** (0.023) 0.089*** (0.023) 
Tree-crops capital   0.077*** (0.010) 0.078*** (0.010) 
Grain-crops capital   0.116*** (0.024) 0.076*** (0.025) 
Animal production capital   0.189*** (0.037) 0.176*** (0.037) 
Constant 1.024*** (0.275) 1.004*** (0.266) 5.128*** (0.334) 4.840*** (0.324) 
Number of obs. 793 793 793 793 
F-statistic 2286.1 2314.6 723.9 731.3 
R-squared 0.924 0.926 0.915 0.909 
Wald test for constant returns to scale     
 F-statistic 12.8 12.5 3.7 0.9 
Prob > F 0.000 0.000 0.053 0.353 
Sum of Coef. 1.055 1.053 1.033°° 1.015°°° 
Dep. var.: agricultural output OLS (1) OLS (2) OLS (3) OLS (4) 
Capital 0.775*** (0.026) 0.733*** (0.029)   
Labor (families) 0.232*** (0.035)  0.446*** (0.035)  
Labor (individuals)  0.264*** (0.034)  0.420*** (0.035) 
Land 0.049** (0.023) 0.056*** (0.022) 0.111*** (0.030) 0.162*** (0.030) 
Reclamation   0.005*** (0.002) 0.005*** (0.002) 
Irrigation   0.009*** (0.002) 0.010*** (0.002) 
Working capital   0.079*** (0.023) 0.089*** (0.023) 
Tree-crops capital   0.077*** (0.010) 0.078*** (0.010) 
Grain-crops capital   0.116*** (0.024) 0.076*** (0.025) 
Animal production capital   0.189*** (0.037) 0.176*** (0.037) 
Constant 1.024*** (0.275) 1.004*** (0.266) 5.128*** (0.334) 4.840*** (0.324) 
Number of obs. 793 793 793 793 
F-statistic 2286.1 2314.6 723.9 731.3 
R-squared 0.924 0.926 0.915 0.909 
Wald test for constant returns to scale     
 F-statistic 12.8 12.5 3.7 0.9 
Prob > F 0.000 0.000 0.053 0.353 
Sum of Coef. 1.055 1.053 1.033°° 1.015°°° 

Notes: Robust standard errors in parenthesis. *Significant at 10 percent, **significant at 5 percent, and ***significant at 1 percent. °Sum of coefficients not statistically different from 1 at 10 percent, °°at 5 percent and °°° at 1 percent.

The sum of the coefficients is very close to one in every specification, but constant returns to scale are statistically confirmed only when the different components of capital are separately entered as regressors (pointing out the limits in the substitutability among some capital items), and thus the results in columns (3) and (4) are preferred to those in columns (1) and (2). The coefficient of land in regression (4) is very close to the share of national rents in output approximately in 1938 (15.6 percent) as estimated by government officials immediately before World War II for fiscal purposes, and thus the coefficients obtained in column (4) are considered a first best. Map 3 represents the normalized TFP levels at the agrarian zone level, estimated from these coefficients.

Map 3.

Estimate of the agricultural total factor productivities of the 793 Italian agricultures around 1930.

Map 3.

Estimate of the agricultural total factor productivities of the 793 Italian agricultures around 1930.

The resulting TFP estimates should be interpreted with caution, as they have been produced from a single cross section, though based on a huge sample. Moreover, some capital items may have been poorly measured or even missed. Nonetheless, if this brief incursion into the realms of TFP is to suggest something, it is that technological differences were not driving the interwar divergence: while there were huge differences across the country in terms of productive efficiency, nothing as a North–South gradient can be identified at the present stage of research. As a whole, Southern agriculture was not more inefficient than the Northern one.

While it is not possible yet to say anything about the relative trends of TFP, a technological divergence would have implied even higher levels of Southern efficiency at the beginning of the century, an implausible hypothesis inconsistent with the tentative estimates for 1911 at the regional level made by Federico (2007). Technological convergence is more likely, but then the Northern success may not be accounted for by faster TFP growth. While further research on the time series of agricultural output and capital is desirable and may reveal something more, the available cross section estimate reinforces the need for a factor-based mechanism for the agricultural divergence. This is precisely what a Von Thünen explanation offers.

Analysis: testing a Von Thünen model for Italian interwar agriculture

This new database allows testing whether a Von Thünen model can satisfactorily account for the spatial patterns of Italian interwar agriculture. I first test the impact of access to markets on output per hectare, as suggested by equation (9). The econometric specification for such a test takes the form:  

(11)
$$\ln y = a + b \ln \hbox{A}{\hbox{M}_n} + x^{\prime}\delta $$

The dependent variable in equation (11) is the logarithm of the agricultural output per hectare, and x′ is a vector of control variables. What exactly is to be empirically considered as representing the “central urban market for agricultural products” of the model is not unambiguous.4 Hence, I consider several alternative definitions of Access to Markets: ln AMnis the logarithm of its n-th definition. I begin the analysis with the simplest, broadest and roughest definition of the demand source: total population. In the real world there is more than one city. A customary way to capture the plurality of demand foci is to build up a measure of access to markets for a given location as a weighted average, where the weights are given by the respective distances to all other locations.

The usual definition of a measure of access to markets takes the form of a spatially weighted sum of the form:  

(12)
$$\hbox{A}{\hbox{M}_i} = \mathop \sum \limits_{\,j \ne i} \displaystyle{{\hbox{Populatio}{\hbox{n}_j}} \over {\hbox{Distanc}{\hbox{e}_j}}} + 2\displaystyle{{\hbox{Populatio}{\hbox{n}_i}} \over {\sqrt {\hbox{Are}{\hbox{a}_i}/\pi } }}$$
where the subscript i denotes the i-th location (agrarian zone), for which the access to markets is to be measured. The variable distance measures the air distance between the gravity center5 of any pair of agrarian zones. Access to markets is defined as the sum of two components: the access to markets other than the considered location itself and the access of such a location to its own market. It is assumed that every location's own population is located halfway between the center of the location's center of gravity and its imaginary external boundary (defined as a circle centered around that center of gravity, with an area equal to that of the given location).

The results are displayed in table 2. Column 1 demonstrates that the coefficient of access to markets measured by the access to total population (according to the 1931 Population Census) is positive and statistically significant, as expected. Being an elasticity, the coefficient points to a 1.32 percent increase in output per hectare as a consequence of a 1 percent increase in the access to markets. In the other columns of table 2 I turn to increasingly restrictive definitions of the sources of demand. In column 2 only the population living agglomerated (i.e., excluding scattered houses) is used in measuring the access to markets. Column 3 reports the results obtained by considering only the agglomerated population living in centers with >10,000 inhabitants. The results are similar in every case. However, a further caveat may be raised against all these measures: the population engaged in agriculture is included therein, and thus suppliers and demanders are counted altogether. This may bias the estimates.

Table 2.

Access to markets and agricultural output—alternative definitions of access to markets

Dependent variable: agricultural output per hectare (1) OLS (2) OLS (3) OLS (4) OLS (5) OLS (6) OLS 
Access to Markets 1 (total population) 1.324*** (0.069)      
Access to Markets 2 (agglomerated population)  1.503*** (0.073)     
Access to Markets 3 (Agg. pop. > 10,000)   1.461*** (0.067)    
Access to Markets 4 (nonagricultural families)    1.106*** (0.055)   
Access to Markets 5 (nonagricultural individuals)     1.158*** (0.058) 1.809*** (0.131) 
Regional dummies No No No No No Yes 
Constant −8.684 −10.391 −8.831 −3.791 −5.867 −13.158 
 0.838*** 0.869*** 0.745*** 0.564*** 0.668*** 1.491*** 
Number of obs. 793 793 793 793 793 793 
F-statistic 372.33 423.57 478.19 398.32 399.62 57.40 
R-squared 0.328 0.350 0.401 0.319 0.329 0.511 
Dependent variable: agricultural output per hectare (1) OLS (2) OLS (3) OLS (4) OLS (5) OLS (6) OLS 
Access to Markets 1 (total population) 1.324*** (0.069)      
Access to Markets 2 (agglomerated population)  1.503*** (0.073)     
Access to Markets 3 (Agg. pop. > 10,000)   1.461*** (0.067)    
Access to Markets 4 (nonagricultural families)    1.106*** (0.055)   
Access to Markets 5 (nonagricultural individuals)     1.158*** (0.058) 1.809*** (0.131) 
Regional dummies No No No No No Yes 
Constant −8.684 −10.391 −8.831 −3.791 −5.867 −13.158 
 0.838*** 0.869*** 0.745*** 0.564*** 0.668*** 1.491*** 
Number of obs. 793 793 793 793 793 793 
F-statistic 372.33 423.57 478.19 398.32 399.62 57.40 
R-squared 0.328 0.350 0.401 0.319 0.329 0.511 

Notes: Robust standard errors in parenthesis; *significant at 10 percent, **significant at 5 percent, and ***significant at 1 percent. All variables in logarithms, when necessary transformed as ln(1 + x).

To avoid this problem, equation (11) is tested considering two measures of the nonagricultural population, which surely exclusively represents the demand side of markets for agricultural products: the first one considering access to families whose family head was not employed in agriculture and the second one considering access to the number of members of such families. I consider the last definition as the preferred measure of access to markets for agricultural products. It is represented in Map 4.

Map 4.

Access to markets for agricultural products in 1931. Access to non-agricultural employment defined as a spatially weighted average of individuals belonging to families whose family head was not employed in agriculture.

Map 4.

Access to markets for agricultural products in 1931. Access to non-agricultural employment defined as a spatially weighted average of individuals belonging to families whose family head was not employed in agriculture.

Columns 4 and 5 report the respective results. The coefficients are slightly reduced with respect to the previous three cases, most likely reflecting the deletion of an undesired effect of the relevant agricultural labor input in the large municipalities on the agricultural output. However, even the effect of these more restrictive definitions of the access to markets remains high and of a comparable order of magnitude, with an elasticity slightly above unity. Moreover, in all cases the access to markets seems to be a main driver of agricultural output: according to the R2 statistic, this variable alone explains between 31 and 54 percent of the variation of output per hectare throughout Italy. This is a remarkable result as it has been obtained without ever considering any other physical, economic, or social element relevant for agricultural production other than access to markets.

However, these results may be due to a generic North–South gradient or to the regional characteristics correlated to the access to markets. In order to test this possibility, column 6 reports the results of a regression with the same definition of the access to markets used in column 5 with a set of regional dummies as additional explicative variables. The coefficient of access to markets actually increases by approximately 50 percent and maintains its high statistical significance, suggesting that its effects on output were not merely a fact of inter-regional (or, more simply, North–South) differences but that it actually explains intra-regional variations of the output per hectare.

Robustness checks

The initial results suggest that a Von Thünen model is worth exploring as an explicative framework for Italian agriculture. Nevertheless, it is unlikely that the agricultural output was exclusively driven by the access to markets. Further robustness checks are required before we can conclude that the access to markets actually shaped Italian agriculture, because its effects may vanish once we consider the impact of differences in the physical environment or socio-economic variables.

The Von Thünen model is built upon the assumption of a featureless plain around a central city, and the geography of Italy is very far from anything like that. Physical characteristics may strongly constrain agricultural potential and may reflect differences in the shifter parameter A of the production function. First, I include as a control the average altitude of each agrarian zone, obtained as an area-weighted average of the altitude of its constituent municipalities (available from ninety-five provincial volumes of the 1929 Agrarian Cadastre). From the same source, I also compute a measure of the slope of the terrain, using information on the absolute maximum and minimum altitude as well as the maximum and minimum altitudes within which the main part of every single municipality was situated. Malaria was still an important disease in some areas of the country, despite the efforts to eradicate it. Thus, from ISTAT (1938), I construct a variable that measures the share of the agrarian zone's land represented by the municipalities where malaria had ever been declared endemic (a feature which may diminish the agricultural potential of a given area).

The way population is weighted (inverse distance weighting) implies that economic distances are proportional to air distances. This is obviously an oversimplification, but I claim that this procedure is reasonable in the present context and that, with all its shortcomings, it does not qualitatively qualify the results. First, the land distances may be reshaped by differential access to transportation means as the roads and railways. Because the transportation infrastructure was particularly dense in the Northern part of the country (particularly between Milan and Turin), obviating it actually results in an underestimation of the access to markets in North-West Italy, which would be increased by a proper measurement of the terrestrial distances. A second problem is that in a country such as Italy, which has a great deal of coastal areas, the geodesic distance may not measure the economic distance well if differences in transport costs between the relevant alternatives (for example, railways and ships) were significant. I deal with this possibility by introducing a control variable of distance to sea in order to correct any measurement distortions attributable to the differences in transport costs. Moreover, I include a dummy variable for agrarian zones that were islands.

Finally, the rainfall regime is a key determinant of the agricultural possibilities of any area. Data on monthly rainfall and on the number of rainy days for 4,632 climatic stations all around the country are available from a series of publications (Annali idrologici) published by several semi-regional authorities working under the Ministry of Public Works (Ministero dei Lavori Pubblici 1936–1939). From this source, I obtained a database consisting of over 400,000 data points, which allows one to fully reproduce the Italian rainfall regime during the thirties by averaging the data on the available 4 years (1936–39) regarding fourteen variables: the average total yearly rainfall and its coefficient of variation, the average seasonal rainfall and their coefficients of variation and the average seasonal rain intensity (measured as the amount of rainfall divided by the number of rainy days). The last available geo-physical variable is latitude, which affects the length of the growing season together with the altitude and the rainfall regime (already taken into account).

I include socio-economic controls along with the physical and geographical ones. The first set of such controls regards the agrarian institutions, the longstanding bête noire of traditional Italian historians (see Cohen 1976; Cohen and Galassi 1990; Federico 2005, 2009; Cohen and Federico 2001). First, differences in the prevailing agrarian contracts are taken into account by including the share of the labor force employed in agriculture being an owner operator, a rented tenant or a sharecropper.6 Second, potential effects of the inequality (Vollrath 2007) in operational sizes and of the inequality in landownership (adjusted by value) are accounted for. The first variable is measured by a Gini index of inequality computed over the farm-size distribution reported by the 1930 Agricultural Census. The second variable is measured by a Gini index computed over the rent-distribution (valued at 1937–39 prices) of actual ownerships, resulting from the merging of all plots and farms belonging to the same owner within an agrarian zone, as reported in a massive official inquiry (INEA 1946–1948) carried out at the end of World War II and reflecting the distribution of the value of land as an asset at the end of the 1930s. From the same source, I take the average rent per property. The latter is a measure of the average value-adjusted size of the ownerships of a given zone, which may have been relevant if there were economies of scale of any type. Finally, the share of land belonging to collective entities (mainly public or semi-public) is included as a control for any potential difference between private and collective landownership management. The source is always INEA (1946–1948).

Another set of socio-economic controls regards human capital accumulation and socio-demographic measures. The only measures of human capital available at the agrarian zone level are the literacy rate and the gap in the literacy rate between females and males (measured as the former rate over the latter), from the 1931 Population Census. From the same source it is possible to derive the female–male ratio, which may have been a proxy of emigration/immigration intensities, given the characteristics of the migrations during this period. The share of the population older than 10, reported in the Census as well, is used as a very rough proxy of birth rates. Another measure of the age structure, which is the only sector-specific available, is the average size of families whose head was employed in agriculture. Finally, the last socio-economic variable is the share of the dispersed population, i.e., the share of the population living in isolated dwellings in the countryside.

Table 3 reports the results of the successive implementation of all the controls. Access to markets is defined, following the most restrictive definition, as the same variable used in the regressions reported in columns 5 and 6 of table 2, namely the access to inhabitants depending on economic activities other than agriculture.7 First, the role of agrarian institutions is assessed. All variables, with the exception of asset-inequality (not significant) and value-adjusted size (positive and significant at 10 percent), have a negative and statistically significant effect on output per hectare. However encouraging these first results may be for traditional historians, the implementation of further controls almost totally erases the negative effects of the agrarian institutions, with the exception of the share of owner operators and of land belonging to collective entities, whose coefficients shrink anyway with respect to the initial estimation. Similarly, no socio-demographic variable is significant at conventional levels after physical variables are taken into account, with the exception of the female–male ratio, with a weak significance at 10 percent.

Table 3.

Access to markets and agricultural output—robustness checks

Dependent variable: agricultural output per hectare OLS (1) OLS (2) OLS (3) OLS (4) OLS (5) 
Access to markets (nonagricultural individuals) 1.158*** (0.058) 1.088*** (0.059) 1.175*** (0.083) 1.231*** (0.086) 1.154*** (0.092) 
Owner operators  −1.456*** (0.269) −1.038*** (0.328) −0.897*** (0.303) −0.878*** (0.280) 
Rented tenants  −0.759*** (0.252) −0.438 (0.301) −0.414 (0.264) −0.152 (0.278) 
Sharecroppers  −0.558*** (0.137) −0.244 (0.229) −0.352* (0.203) −0.137 (0.203) 
Gini of farms (size)  −1.362*** (0.344) −1.570*** (0.396) −0.956** (0.373) −0.574 (0.366) 
Gini of private ownerships (value)  −0.743 (0.689) −0.422 (0.779) −0.341 (0.606) −0.905 (0.573) 
Average rent per ownership  0.050* (0.028) 0.086*** (0.032) 0.082*** (0.030) 0.045 (0.029) 
Share of land of collective entities  −1.211*** (0.171) −0.958*** (0.188) −0.733*** (0.204) −0.786*** (0.225) 
Literacy rate   −1.108** (0.478) 0.180 (0.475) 0.283 (0.481) 
Female literacy rate/male literacy rate   0.693 (0.540) −0.637 (0.549) 0.022 (0.571) 
Female–male ratio   0.389 (0.400) 0.986*** (0.343) 0.631* (0.337) 
Share of inhabitants > 10 years old   1.841 (1.688) 1.086 (1.194) 0.650 (1.058) 
Size of agricultural families   −0.348* (0.202) −0.297 (0.180) −0.303 (0.195) 
Share of dispersed population   −0.077 (0.157) 0.033 (0.137) 0.079 (0.138) 
Altitude    −0.001*** (0.000) −0.001*** (0.000) 
Terrain ruggedness    0.001*** (0.000) 0.001*** (0.000) 
Share of malaria endemic area    −0.335*** (0.078) −0.283*** (0.077) 
Distance to the Sea    −0.049*** (0.013) −0.042*** (0.014) 
Island dummy    0.575* (0.313) 0.314 (0.316) 
Latitude (km from Equator)    −3.717*** (0.841) −4.889*** (1.164) 
Total rainfall     0.943 (0.661) 
CV total rainfall     0.201 (0.452) 
Winter rainfall     −0.484** (0.202) 
Spring rainfall     −0.737*** (0.265) 
Summer rainfall     0.146 (0.144) 
Autumn rainfall     0.334 (0.336) 
CV winter rainfall     −0.156 (0.212) 
CV spring rainfall     −0.254 (0.313) 
CV summer rainfall     −0.217 (0.156) 
CV autumn rainfall     0.227 (0.246) 
Winter rain Intensity     0.454** (0.201) 
Spring rain intensity     −0.120 (0.213) 
Summer rain intensity     −0.267** (0.133) 
Autumn rain intensity     −0.380* (0.220) 
North dummy   0.011 (0.077) 0.238*** (0.069) 0.137* (0.079) 
South dummy   −0.009 (0.084) −0.115 (0.082) −0.146* (0.082) 
Constant −5.867*** (0.668) −3.454*** (0.596) −5.587*** (1.675) 25.136*** (7.088) 34.920*** (10.015) 
Number of obs. 793 732 732 732 732 
F-statistic 399.62 159.87 93.38 90.74 68.42 
R-squared 0.329 0.621 0.632 0.720 0.752 
Dependent variable: agricultural output per hectare OLS (1) OLS (2) OLS (3) OLS (4) OLS (5) 
Access to markets (nonagricultural individuals) 1.158*** (0.058) 1.088*** (0.059) 1.175*** (0.083) 1.231*** (0.086) 1.154*** (0.092) 
Owner operators  −1.456*** (0.269) −1.038*** (0.328) −0.897*** (0.303) −0.878*** (0.280) 
Rented tenants  −0.759*** (0.252) −0.438 (0.301) −0.414 (0.264) −0.152 (0.278) 
Sharecroppers  −0.558*** (0.137) −0.244 (0.229) −0.352* (0.203) −0.137 (0.203) 
Gini of farms (size)  −1.362*** (0.344) −1.570*** (0.396) −0.956** (0.373) −0.574 (0.366) 
Gini of private ownerships (value)  −0.743 (0.689) −0.422 (0.779) −0.341 (0.606) −0.905 (0.573) 
Average rent per ownership  0.050* (0.028) 0.086*** (0.032) 0.082*** (0.030) 0.045 (0.029) 
Share of land of collective entities  −1.211*** (0.171) −0.958*** (0.188) −0.733*** (0.204) −0.786*** (0.225) 
Literacy rate   −1.108** (0.478) 0.180 (0.475) 0.283 (0.481) 
Female literacy rate/male literacy rate   0.693 (0.540) −0.637 (0.549) 0.022 (0.571) 
Female–male ratio   0.389 (0.400) 0.986*** (0.343) 0.631* (0.337) 
Share of inhabitants > 10 years old   1.841 (1.688) 1.086 (1.194) 0.650 (1.058) 
Size of agricultural families   −0.348* (0.202) −0.297 (0.180) −0.303 (0.195) 
Share of dispersed population   −0.077 (0.157) 0.033 (0.137) 0.079 (0.138) 
Altitude    −0.001*** (0.000) −0.001*** (0.000) 
Terrain ruggedness    0.001*** (0.000) 0.001*** (0.000) 
Share of malaria endemic area    −0.335*** (0.078) −0.283*** (0.077) 
Distance to the Sea    −0.049*** (0.013) −0.042*** (0.014) 
Island dummy    0.575* (0.313) 0.314 (0.316) 
Latitude (km from Equator)    −3.717*** (0.841) −4.889*** (1.164) 
Total rainfall     0.943 (0.661) 
CV total rainfall     0.201 (0.452) 
Winter rainfall     −0.484** (0.202) 
Spring rainfall     −0.737*** (0.265) 
Summer rainfall     0.146 (0.144) 
Autumn rainfall     0.334 (0.336) 
CV winter rainfall     −0.156 (0.212) 
CV spring rainfall     −0.254 (0.313) 
CV summer rainfall     −0.217 (0.156) 
CV autumn rainfall     0.227 (0.246) 
Winter rain Intensity     0.454** (0.201) 
Spring rain intensity     −0.120 (0.213) 
Summer rain intensity     −0.267** (0.133) 
Autumn rain intensity     −0.380* (0.220) 
North dummy   0.011 (0.077) 0.238*** (0.069) 0.137* (0.079) 
South dummy   −0.009 (0.084) −0.115 (0.082) −0.146* (0.082) 
Constant −5.867*** (0.668) −3.454*** (0.596) −5.587*** (1.675) 25.136*** (7.088) 34.920*** (10.015) 
Number of obs. 793 732 732 732 732 
F-statistic 399.62 159.87 93.38 90.74 68.42 
R-squared 0.329 0.621 0.632 0.720 0.752 

Notes: Robust standard errors in parenthesis; *significant at 10 percent, **significant at 5 percent, and ***significant at 1 percent. All variables in logarithms, when necessary transformed as ln(1 + x).

What the results of table 3 suggest is that most of the socio-economic features, apparently driving agricultural performance, were ultimately determined by the environmental features (with the notable exception of the ownership patterns). Indeed, the majority of variables with explicative power are of a physical nature: as expected, the altitude, malaria potential, and distance to the sea had a negative impact on the agricultural output, as the latitude had. Other variables display more puzzling results, as a positive impact of ruggedness. Possibly this result reflects the fact that, once average altitude is taken into account, a certain range of variation of altitude within a given agrarian zone allows for the positive complementarities. High quality vineyards were often found on hill slopes, rather than in the plains, and this effect can largely account for the sign of the coefficient. In addition, many rainfall regime variables are statistically significant.

A remarkable result in this analysis is that, despite the strong explicative power of physical variables, the coefficient of access to markets is hardly changed, even after the inclusion of more than thirty control variables. Its statistical significance remains high, consistently at 1 percent. The elasticity of output with respect to the access to markets fluctuates in all specifications around values slightly above unity, which are in a surprisingly reduced range (between 1.088 and 1.231). This fact is worth stressing because few “human originated” variables resist the introduction of “nature made” variables into the analysis. In summation, table 3 strongly confirms the existence of a Von Thünen pattern in Italian interwar agriculture.

The Von Thünen model described in Section 3 not only predicts a gradient of output intensity around foci of demand sources but also predicts that this will be generated by an input intensity gradient and will result in another gradient in rents per hectare. Using the available data, it is possible to test all these hypotheses as well and, hence, to obtain stronger evidence of the accuracy of the theory as a description of the real world. If four different predictions derived from the same theoretical model are found consistent with reality, these findings will result in a strong confirmation of the soundness of the model.

Hence, in table 4, I report the results of estimating equations (13), (14), and (15) with the available database.  

(13)
$$ \; \ln l = c + d\ \!\! \ln \hbox{A}{\hbox{M}_n} + x^{\prime}\delta $$
 
(14)
$$\ln k = e + f\ \!\! \ln \hbox{A}{\hbox{M}_n} + x^{\prime}\delta $$
 
(15)
$$\ln R = g + h\ \!\! \ln \hbox{A}{\hbox{M}_n} + x^{\prime}\delta $$

Table 4.

The impact of access to markets on factor intensity and land rents

Dependent variable Agricultural labor per hectare (families)
 
Agricultural labor per hectare (family members)
 
Agricultural capital per hectare
 
Rent per hectare
 
OLS (1) OLS (2) OLS (3) OLS (4) OLS (5) OLS (6) OLS (7) OLS (8) 
Access to markets (nonagricultural individuals) 0.910*** (0.060) 1.243*** (0.103) 1.088*** (0.059) 1.247*** (0.103) 1.097*** (0.055) 1.042*** (0.099) 1.314*** (0.093) 0.938*** (0.111) 
Owner-operators  −0.959*** (0.309)  −0.950*** (0.309)  −0.803*** (0.309)  −0.448 (0.406) 
Rented tenants  0.000 (0.278)  −0.006 (0.278)  −0.187 (0.298)  −0.072 (0.307) 
Sharecroppers  −0.031 (0.203)  −0.036 (0.203)  −0.188 (0.210)  −0.485** (0.235) 
Gini of farms (size)  −1.055*** (0.405)  −1.059*** (0.405)  −0.079 (0.360)  −0.750 (0.573) 
Gini of private ownerships (value)  −2.899*** (0.584)  −2.890*** (0.584)  −1.196* (0.635)  −3.754*** (0.727) 
Average rent per ownership  −0.159*** (0.037)  −0.158*** (0.037)  −0.045 (0.033)  0.389*** (0.044) 
Share of land of collective entities  −0.669*** (0.201)  −0.666*** (0.202)  −0.533** (0.217)  −0.869*** (0.227) 
Literacy rate  −0.317 (0.542)  −0.316 (0.542)  1.432** (0.614)  1.802*** (0.571) 
Literacy rate Gap  −0.136 (0.609)  −0.136 (0.609)  0.101 (0.650)  −0.604 (0.646) 
Female–male Ratio  0.736** (0.323)  0.730** (0.324)  0.017 (0.394)  0.781* (0.472) 
Share of inhabitants > 10 years old  −1.552 (0.985)  −1.562 (0.987)  0.463 (1.050)  −1.066 (1.174) 
Size of agricultural families  −0.560*** (0.198)  0.636*** (0.198)  −0.071 (0.202)  −0.255 (0.227) 
Share of dispersed population  −0.097 (0.146)  −0.096 (0.146)  0.194 (0.149)  −0.416** (0.167) 
Altitude  −0.001*** (0.000)  −0.001*** (0.000)  −0.001*** (0.000)  −0.001*** (0.000) 
Terrain ruggedness  0.001*** (0.000)  0.001*** (0.000)  0.001*** (0.000)  0.000 (0.000) 
Share of malaria endemic area  −0.328*** (0.086)  −0.328*** (0.086)  −0.205** (0.082)  −0.277*** (0.079) 
Distance to the sea  −0.039*** (0.013)  −0.038*** (0.013)  −0.045*** (0.014)  −0.011 (0.014) 
Island dummy  0.158 (0.244)  0.156 (0.244)  0.063 (0.361)  −0.578** (0.260) 
Latitude (km from Equator)  −2.548** (1.099)  −2.539** (1.099)  −0.084 (1.464)  −6.098*** (1.324) 
Rainfall regime (14 vars.) No Yes No Yes No Yes No Yes 
North–south dummies No Yes No Yes No Yes No Yes 
Constant −12.058*** (0.694) 7.178 (9.446) −12.486*** (0.684) 6.524 (9.445) −4.558*** (0.633) −4.510 (12.817) −9.661*** (1.065) 44.943*** (11.369) 
Number of obs. 793 732 793 732 793 732 733 732 
F-statistic 228.1 38.7 334.9 52.4 397.4 49.5 200.3 135.2 
R-squared 0.249 0.687 0.324 0.717 0.316 0.679 0.195 0.863 
Dependent variable Agricultural labor per hectare (families)
 
Agricultural labor per hectare (family members)
 
Agricultural capital per hectare
 
Rent per hectare
 
OLS (1) OLS (2) OLS (3) OLS (4) OLS (5) OLS (6) OLS (7) OLS (8) 
Access to markets (nonagricultural individuals) 0.910*** (0.060) 1.243*** (0.103) 1.088*** (0.059) 1.247*** (0.103) 1.097*** (0.055) 1.042*** (0.099) 1.314*** (0.093) 0.938*** (0.111) 
Owner-operators  −0.959*** (0.309)  −0.950*** (0.309)  −0.803*** (0.309)  −0.448 (0.406) 
Rented tenants  0.000 (0.278)  −0.006 (0.278)  −0.187 (0.298)  −0.072 (0.307) 
Sharecroppers  −0.031 (0.203)  −0.036 (0.203)  −0.188 (0.210)  −0.485** (0.235) 
Gini of farms (size)  −1.055*** (0.405)  −1.059*** (0.405)  −0.079 (0.360)  −0.750 (0.573) 
Gini of private ownerships (value)  −2.899*** (0.584)  −2.890*** (0.584)  −1.196* (0.635)  −3.754*** (0.727) 
Average rent per ownership  −0.159*** (0.037)  −0.158*** (0.037)  −0.045 (0.033)  0.389*** (0.044) 
Share of land of collective entities  −0.669*** (0.201)  −0.666*** (0.202)  −0.533** (0.217)  −0.869*** (0.227) 
Literacy rate  −0.317 (0.542)  −0.316 (0.542)  1.432** (0.614)  1.802*** (0.571) 
Literacy rate Gap  −0.136 (0.609)  −0.136 (0.609)  0.101 (0.650)  −0.604 (0.646) 
Female–male Ratio  0.736** (0.323)  0.730** (0.324)  0.017 (0.394)  0.781* (0.472) 
Share of inhabitants > 10 years old  −1.552 (0.985)  −1.562 (0.987)  0.463 (1.050)  −1.066 (1.174) 
Size of agricultural families  −0.560*** (0.198)  0.636*** (0.198)  −0.071 (0.202)  −0.255 (0.227) 
Share of dispersed population  −0.097 (0.146)  −0.096 (0.146)  0.194 (0.149)  −0.416** (0.167) 
Altitude  −0.001*** (0.000)  −0.001*** (0.000)  −0.001*** (0.000)  −0.001*** (0.000) 
Terrain ruggedness  0.001*** (0.000)  0.001*** (0.000)  0.001*** (0.000)  0.000 (0.000) 
Share of malaria endemic area  −0.328*** (0.086)  −0.328*** (0.086)  −0.205** (0.082)  −0.277*** (0.079) 
Distance to the sea  −0.039*** (0.013)  −0.038*** (0.013)  −0.045*** (0.014)  −0.011 (0.014) 
Island dummy  0.158 (0.244)  0.156 (0.244)  0.063 (0.361)  −0.578** (0.260) 
Latitude (km from Equator)  −2.548** (1.099)  −2.539** (1.099)  −0.084 (1.464)  −6.098*** (1.324) 
Rainfall regime (14 vars.) No Yes No Yes No Yes No Yes 
North–south dummies No Yes No Yes No Yes No Yes 
Constant −12.058*** (0.694) 7.178 (9.446) −12.486*** (0.684) 6.524 (9.445) −4.558*** (0.633) −4.510 (12.817) −9.661*** (1.065) 44.943*** (11.369) 
Number of obs. 793 732 793 732 793 732 733 732 
F-statistic 228.1 38.7 334.9 52.4 397.4 49.5 200.3 135.2 
R-squared 0.249 0.687 0.324 0.717 0.316 0.679 0.195 0.863 

Notes: Robust standard errors in parenthesis; *significant at 10 percent, **significant at 5 percent, and ***significant at 1 percent. All variables in logarithms, when necessary transformed as ln(1 + x).

Each variable is first regressed exclusively on access to markets, and then the same full set of control variables used in column 5 of table 3 is introduced. As stated, employment in agriculture at the agrarian zone level is not available for 1931, so I use as the dependent variables both the (log of the) number of families whose family head was employed in agriculture per hectare, which proxies the lower bound of agricultural employment, and the number of members of such families per hectare, which represents the upper bound.

The results are very similar in both cases with respect to access to markets (the variable of interest here). Indeed, results are very similar when we consider the impact of access to markets on either labor input per hectare, capital input per hectare or rents per hectare. As predicted by the model, the higher access to markets results in higher labor and capital inputs per hectare and higher rents. Access to markets alone explains close to one-third of the variation of output and capital per hectare, as well as between one-quarter and one-third of the variation in labor per hectare and one-fifth of the variation of rents in the whole sample. The elasticity is in all cases close to unity, and noticeably remains rather unchanged (as well as its statistical significance) even after the inclusion of a large set of controls.

The results displayed in table 4 strongly support the inference already derived from table 3. The divergence of Italian agricultures in the first half of the twentieth century may very well be explained by the growth of the nonagricultural sector in Northern Italy (and especially in the Northwestern region), because the whole pattern of the sector suggests that it reacted positively to the geographic access to sources of demand in sectors other than agriculture. According to this framework, greater access to markets resulted in a higher intensity in the use of productive factors (other than land) per unit of land.

Table 4 has indicated that the data support this channel. However, it is possible to directly explore the mechanisms through which access to markets shaped agricultural activity. It is easy to see that the equilibrium output per hectare at a location d may be expressed as:  

(16)
$$\ln {y_d}^* = \ln A + \alpha \ln {k_d}^* + \beta \ln {l_d}^* $$

That is, access to markets shapes output per hectare by shaping the factor intensities. We can explicitly estimate (16) in order to check such a mechanism. The results are displayed in table 5. According to equation (16), once we regress output per hectare on capital per hectare, there should still be an effect of the access to markets, which determines labor per hectare as well. Vice versa, when regressing output on labor, access to markets must have a positive effect as it determined, in addition to labor inputs, the unaccounted for capital input. Indeed, statistical results confirm this mechanism.

Table 5.

The effect of access to markets on agricultural output—mechanisms

Dependent variable: agricultural output per hectare OLS (1) OLS (2) OLS (3) OLS (4) 
Access to markets 1.154*** (0.092) 0.350*** (0.093) 0.455*** (0.077) 0.292*** (0.077) 
Agricultural capital per hectare   0.672*** (0.029) 0.493*** (0.041) 
Agricultural labor per hectare  0.645*** (0.047)  0.280*** (0.051) 
Agrarian institutions (7 variables) Yes Yes Yes Yes 
Socio-demographic variables (6 variables) Yes Yes Yes Yes 
Physical variables (6 variables) Yes Yes Yes Yes 
Rainfall regime (14 variables) Yes Yes Yes Yes 
North–south dummies Yes Yes Yes Yes 
Constant 34.920*** (10.015) 30.710*** (8.049) 37.949*** (6.566) 35.319*** (6.284) 
Number of obs. 732 732 732 732 
F-statistic 68.4 147.5 172.6 203.5 
R-squared 0.752 0.858 0.886 0.896 
Dependent variable: agricultural output per hectare OLS (1) OLS (2) OLS (3) OLS (4) 
Access to markets 1.154*** (0.092) 0.350*** (0.093) 0.455*** (0.077) 0.292*** (0.077) 
Agricultural capital per hectare   0.672*** (0.029) 0.493*** (0.041) 
Agricultural labor per hectare  0.645*** (0.047)  0.280*** (0.051) 
Agrarian institutions (7 variables) Yes Yes Yes Yes 
Socio-demographic variables (6 variables) Yes Yes Yes Yes 
Physical variables (6 variables) Yes Yes Yes Yes 
Rainfall regime (14 variables) Yes Yes Yes Yes 
North–south dummies Yes Yes Yes Yes 
Constant 34.920*** (10.015) 30.710*** (8.049) 37.949*** (6.566) 35.319*** (6.284) 
Number of obs. 732 732 732 732 
F-statistic 68.4 147.5 172.6 203.5 
R-squared 0.752 0.858 0.886 0.896 

Notes: Robust standard errors in parenthesis; *significant at 10 percent, **significant at 5 percent, and ***significant at 1 percent. All variables in logarithms, when necessary transformed as ln(1 + x).

When just one input (either capital or labor) is included as a regressor (columns 2 and 3), the coefficient of access to markets shrinks but remains statistically significant. Column 4 indicates that after labor and capital are simultaneously taken into account, the access to markets still has a positive and highly significant effect, despite the magnitude of the coefficient being reduced to a mere 25 percent of its initial value. The latter is an unexpected result, because according to the theory there is no room for the access to markets shaping output after the factor uses are taken into account. A possible explanation for this result is the measurement error: there may have been areas with a high access to markets having certain items of capital unaccounted for in the present estimate. Another explanation may be related to a higher A in the areas with high access to markets: closeness to sources of demand may have provided a better access to information, thus allowing specialization in more risky, though more profitable, products (for example). However, it shall be stressed that the residual quantitative effect of the access to markets is small if compared with its direct effect via factor intensity. Putting it another way, the access to markets strongly determined agricultural production, with elasticity close to unity; 75 percent of its effects are explained by factor use and a residual 25 percent through still unexplained channels.

A further doubt may be raised. Access to markets may determine higher agricultural output, but if agricultural suitability favors the growth of nonagricultural activities, the access to markets may be endogenous. In those cases, the OLS estimation would be biased. Although agricultural suitability may determine nonindustrial growth in a location but not necessarily in its neighborhood (which is what is measured by access to markets), the issue deserves a proper research strategy. A fully exogenous variable is required to estimate (12) by instrumental variables. I use the access to the HP generated by water-powered engines in the 254 main industrial municipalities as reported by the 1911 Industrial Census (Ministero di Agricoltura, Industria e Commercio 1914). This is surely an exogenous variable for two reasons. First, it is related to a previous period, and hence it cannot be determined by the dependent variable. In this sense, it constitutes a proper lag, which is a usual procedure in these contexts. Second, HP generated by water in 1911 in the main industrial municipalities is related mainly to hydroelectric power, the development of which was correlated with the local conditions for the use of waterfalls as a source of electric power and not with the agricultural output; most of these municipalities were in the Alpine belt around Turin and Milan.

The results of IV estimation are displayed in table 6. The first stage reveals that this is a relevant instrument, and the magnitude of the coefficient is roughly similar. The only relevant change with respect to OLS can be found in the exploration of the mechanisms: once labor and capital are taken into account, the (theoretically unexplained) residual effect of access to markets is still positive with a magnitude similar to that estimated by OLS, but it is only significant at the 10 percent level. Indeed, this effect goes in the direction suggested by theory, thus reinforcing it.

Table 6.

Instrumental variables estimation

Dependent variable: agricultural output per hectare IV IV IV IV IV IV IV IV 
First Stage Second Stage First stage Second stage First stage Second stage First stage Second stage 
(1) (2) (3) (4) (5) (6) (7) (8) 
Access to markets  1.146*** (0.188)  0.429** (0.182)  0.371** (0.145)  0.263* (0.151) 
Access to HP by Hydraulic Motors 1911 0.339*** (0.022)  0.283*** (0.020)  0.297*** (0.021)  0.282*** (0.020)  
Agricultural Labor per hectare   0.147*** (0.012) 0.631*** (0.058)   0.139*** (0.017) 0.284*** (0.057) 
Agricultural capital per hectare     0.109*** (0.012) 0.684*** (0.036) 0.011 (0.016) 0.494*** (0.041) 
Agrarian Institutions (7 variables) Yes Yes Yes Yes Yes Yes Yes Yes 
Socio-demographic variables (6 variables) Yes Yes Yes Yes Yes Yes Yes Yes 
Physical variables (6 variables) Yes Yes Yes Yes Yes Yes Yes Yes 
Rainfall regime (14 variables) Yes Yes Yes Yes Yes Yes Yes Yes 
North–south dummies Yes Yes Yes Yes Yes Yes Yes Yes 
Constant 14.751*** (3.454) 34.936*** (10.000) 11.306*** (3.126) 30.689*** (8.066) 13.442*** (3.264) 38.137*** (6.648) 11.365*** (3.129) 35.334*** (6.295) 
Number of obs. 732 732 732 732 732 732 732 732 
F-statistic 115.61 65.96 142.74 144.94 128.51 172.61 138.88 201.37 
R-squared 0.857 0.752 0.884 0.858 0.873 0.886 0.884 0.896 
Dependent variable: agricultural output per hectare IV IV IV IV IV IV IV IV 
First Stage Second Stage First stage Second stage First stage Second stage First stage Second stage 
(1) (2) (3) (4) (5) (6) (7) (8) 
Access to markets  1.146*** (0.188)  0.429** (0.182)  0.371** (0.145)  0.263* (0.151) 
Access to HP by Hydraulic Motors 1911 0.339*** (0.022)  0.283*** (0.020)  0.297*** (0.021)  0.282*** (0.020)  
Agricultural Labor per hectare   0.147*** (0.012) 0.631*** (0.058)   0.139*** (0.017) 0.284*** (0.057) 
Agricultural capital per hectare     0.109*** (0.012) 0.684*** (0.036) 0.011 (0.016) 0.494*** (0.041) 
Agrarian Institutions (7 variables) Yes Yes Yes Yes Yes Yes Yes Yes 
Socio-demographic variables (6 variables) Yes Yes Yes Yes Yes Yes Yes Yes 
Physical variables (6 variables) Yes Yes Yes Yes Yes Yes Yes Yes 
Rainfall regime (14 variables) Yes Yes Yes Yes Yes Yes Yes Yes 
North–south dummies Yes Yes Yes Yes Yes Yes Yes Yes 
Constant 14.751*** (3.454) 34.936*** (10.000) 11.306*** (3.126) 30.689*** (8.066) 13.442*** (3.264) 38.137*** (6.648) 11.365*** (3.129) 35.334*** (6.295) 
Number of obs. 732 732 732 732 732 732 732 732 
F-statistic 115.61 65.96 142.74 144.94 128.51 172.61 138.88 201.37 
R-squared 0.857 0.752 0.884 0.858 0.873 0.886 0.884 0.896 

Notes: Robust standard errors in parenthesis; *significant at 10 percent, **significant at 5 percent, and ***significant at 1 percent. All variables in logarithms, when necessary transformed as ln(1 + x).

In table 7, I implement an additional robustness check. Throughout the paper, the access to markets has been exclusively defined in domestic terms. This approach is justified by the existence of tariffs and by the restrictive trade policy implemented by the Italian government since 1926, which aimed to achieve self-sufficiency in many products. Nonetheless, a fully autarkic policy was implemented only after the Italian invasion of Ethiopia and the consequent sanctions of the League of Nations (Ciocca and Toniolo 1976; Toniolo 1980). Hence, Italy was by no means a fully closed economy around 1929 and, if the demand for agricultural products mattered, the international demand should also be taken into account, especially to check how the results previously found hold to the inclusion of a broader world.

Table 7.

Further robustness checks—domestic and foreign markets

Dependent variable: agricultural output per hectare OLS OLS OLS IV IV IV IV IV IV IV IV 
   1st Stage 2nd Stage 1st Stage 2nd Stage 1st Stage 2nd Stage 1st Stage 2nd Stage 
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) 
Access to Markets (domestic) 1.154*** (0.092) 1.170*** (0.090)   1.108*** (0.187)  0.405** (0.180)  0.375*** (0.143)  0.264* (0.148) 
Access to foreign markets  2.546*** (0.848)  −0.783*** (0.292) 2.517*** (0.854) −0.896*** (0.263) 1.483** (0.629) −1.184*** (0.277) −0.427 (0.594) −0.957*** (0.267) −0.071 (0.555) 
Total access to markets (F + D)   2.139*** (0.185)         
Capital per hectare        0.116*** (0.012) 0.686*** (0.037) 0.021 (0.017) 0.495*** (0.042) 
Labor per hectare      0.148*** (0.011) 0.632*** (0.058)   0.132*** (0.017) 0.284*** (0.057) 
Access to HP by Hydraulic Motors 1911    0.343*** (0.022)  0.287*** (0.020)  0.300*** (0.021)  0.285*** (0.020)  
Agrarian institutions (7 variables) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Socio-demographic variables (6 variables) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Physical variables (6 variables) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Rainfall regime (14 variables) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
North–South dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Constant 34.920*** (10.015) 59.123*** (12.671) 45.065*** (9.927) 7.453* (4.386) 58.962*** (12.705) 2.922 (3.958) 44.834*** (10.510) 2.312*** (4.141) 34.067*** (8.288) 2.469 (3.972) 34.661*** (7.823) 
Number of obs. 732 732 732 732 732 732 732 732 732 732 732 
F-statistic 68.4 68.4 66.5 113.7 64.9 141.4 141.0 128.7 169.8 138.0 197.5 
R-squared 0.752 0.756 0.751 0.858 0.756 0.886 0.860 0.876 0.886 0.886 0.896 
Dependent variable: agricultural output per hectare OLS OLS OLS IV IV IV IV IV IV IV IV 
   1st Stage 2nd Stage 1st Stage 2nd Stage 1st Stage 2nd Stage 1st Stage 2nd Stage 
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) 
Access to Markets (domestic) 1.154*** (0.092) 1.170*** (0.090)   1.108*** (0.187)  0.405** (0.180)  0.375*** (0.143)  0.264* (0.148) 
Access to foreign markets  2.546*** (0.848)  −0.783*** (0.292) 2.517*** (0.854) −0.896*** (0.263) 1.483** (0.629) −1.184*** (0.277) −0.427 (0.594) −0.957*** (0.267) −0.071 (0.555) 
Total access to markets (F + D)   2.139*** (0.185)         
Capital per hectare        0.116*** (0.012) 0.686*** (0.037) 0.021 (0.017) 0.495*** (0.042) 
Labor per hectare      0.148*** (0.011) 0.632*** (0.058)   0.132*** (0.017) 0.284*** (0.057) 
Access to HP by Hydraulic Motors 1911    0.343*** (0.022)  0.287*** (0.020)  0.300*** (0.021)  0.285*** (0.020)  
Agrarian institutions (7 variables) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Socio-demographic variables (6 variables) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Physical variables (6 variables) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Rainfall regime (14 variables) Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
North–South dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 
Constant 34.920*** (10.015) 59.123*** (12.671) 45.065*** (9.927) 7.453* (4.386) 58.962*** (12.705) 2.922 (3.958) 44.834*** (10.510) 2.312*** (4.141) 34.067*** (8.288) 2.469 (3.972) 34.661*** (7.823) 
Number of obs. 732 732 732 732 732 732 732 732 732 732 732 
F-statistic 68.4 68.4 66.5 113.7 64.9 141.4 141.0 128.7 169.8 138.0 197.5 
R-squared 0.752 0.756 0.751 0.858 0.756 0.886 0.860 0.876 0.886 0.886 0.896 

Notes: Robust standard errors in parenthesis; * significant at 10 percent, ** significant at 5 percent, and *** significant at 1 percent. All variables in logarithms, when necessary transformed as ln(1 + x).

The existence of tariffs implied a distinct treatment for every product according to its origin, and this makes the estimation of the effects of international markets complex. However, a rough approximation can be obtained. I measure access to foreign markets as a weighted by distance sum of the population of all cities with more than 100,000 inhabitants in Europe in 1930 and with more than 200,000 inhabitants in 1930 (occasionally in 1935) outside Europe. I collected the data from the German Statistical Yearbook of 1941–42 (Statistisches Reichsamt 1942). The resulting measure points to a higher access to foreign markets in Northern Italy as a whole.

With this measure, all of the relevant regressions have been re-calculated and are displayed in table 7. First, the access to foreign markets is included as an independent variable along with the access to domestic markets (which is the measure used throughout the paper). Access to foreign markets is statistically significant and has a large coefficient. However, the role played by the access to domestic markets remains unchanged, both in magnitude and in significance. In column 3, I also use a measure of total access to markets, summing the domestic and foreign ones, the coefficients of which are positive and significant as well, with a coefficient higher than that of access to the domestic markets alone, as it is reasonable.

When the mechanisms are explored, we find that the effect of the access to foreign markets on agricultural output vanishes once capital is accounted for (unlike the access to domestic markets) but not when labor is individually accounted for. These results seem to suggest that access to foreign markets impacted output through higher inputs of capital but not through higher inputs of labor. Although these facts deserve further investigation, the role played by the access to domestic markets as a key driver in the spatial organization of the agricultural production, as investigated throughout the paper, remains unchanged.

The last robustness check consists in controlling for spatial effects. Although the IV estimates suggest that omitted variables are not qualifying the results, there are other potential effects of spatial dependence which may suggest the implementation of such a control. Spatial autocorrelation of the dependent or of some independent variables may arise, for example, if there were spatial spillovers (as, e.g., when the value of output in a given location raises the demand for output of the neighboring locations). Similar to Carmona and Rosés (2012) in their paper on Spanish land prices, I address the spatial effects by following the methodology suggested by Badinger et al. (2004). The procedure consists in spatially filtering the data, with the spatial filter being based on the measure of spatial dependence proposed by Getis and Ord (1992), G(δ).

Hence, the i-th observation of the variable x is transformed into:  

(17)
$${\tilde x_i} = {x_i}\left[ {\displaystyle{{\mathop \sum \nolimits_{\,j = 1}^n {w_{ij}}(\delta )} \over {({N - 1} ){G_i}(\delta )}}\; } \right],\; \quad i \ne j$$

with Gi(δ) being:  

(18)
$${G_i}(\delta )= \; \displaystyle{{\mathop \sum \nolimits_{\,j = 1}^n {w_{ij}}(\delta ){x_j}} \over {\mathop \sum \nolimits_{\,j = 1}^n {x_j}}}\; \; ,\; \quad i \ne j\; $$
wij(δ) represents the elements of the so-called spatial weight matrix W and depends on a distance decay function depending on the parameter δ. As is usual in this context, I assume that the elements of the spatial weight matrix take the form of:  
(19)
$${w_{ij}}(\delta )= {d_{ij}}^{ - \delta } ,\quad i \ne j$$
with dij being the normalized distance between any pair of observations and δ = 1.

Therefore, it is possible to rewrite the expression of the spatially filtered observation as:  

(20)
$${\tilde x_i} = {x_i}\left[ {\displaystyle{{\mathop \sum \nolimits_{\,j = 1}^n {x_j}} \over {({N - 1} )}}} \right]\left[ {\mathop \sum \limits_{\,j = 1}^n \displaystyle{1 \over {{d_{ij}}}}} \right] {\displaystyle \Bigg[{1 \over {\mathop \sum \nolimits_{\,j = 1}^n {{{x_j}} \over {{d_{ij}}}}}}} \Bigg]\; ,\quad \; i \ne j$$

It is easy to see that a spatially filtered variable is equivalent to the original if (i) all observations are at the same distance from each other and (ii) all observations share the same value of the variable. Conversely, relatively central locations with variable values well above the mean will see the spatially filtered value reduced. The second term of the last equation is simply the average of all observations but the i-th. The third term is a measure of the centrality of a location: the more distant an observation is from all the others, the lower its value. The fourth term is the inverse of the access to the value of the variable available in all other locations.

Intuitively, the spatial filter for a given location can be seen as the interaction of these three elements: the product of the mean of all other locations by the centrality of the location, divided by its access to all other values. If access to the spatially distributed value of a variable from a given location is found to be higher than the (spaceless) average value corrected by a general centrality measure of the location, the high values of the considered variable are more frequently clustered around a given location than if the data were distributed randomly across space. This would be the case if the values of the variable in a given location were spatially dependent upon the value of its neighbors. The spatial filter removes every effect of such spatial dependence.

Table 8 presents the results obtained with spatially filtered variables.8 The results by and large confirm the role of the access to markets in shaping all the variables considered after the removal of spatial effects. Indeed, while the coefficient of access to markets even increases somewhat (with the elasticities slightly above unity), the results are broadly similar to the baseline estimate. When mechanisms are considered, the coefficient of access to markets collapses by an order of magnitude of approximately 75 percent, confirming that shaping factor demands was the main mechanism by which access to markets had an impact on agricultural output. Table 8 also displays the mechanisms by which the other variables affected inputs. A number of institutional variables, notably land inequality (which may have had an effect on the efficiency of factor markets), and distance to the sea, also had a negative impact on output totally accounted for by the reduced use of inputs.

Table 8.

Further robustness checks—controlling for spatial effects

(All variables spatially filtered)
 
Dependent variable Agricultural labor per hectare (families) Agricultural labor per hectare (family members) Agricultural capital per hectare Rent per hectare Agricultural output per hectare
 
 OLS OLS OLS OLS OLS OLS IV IV 
       1st Stage 2nd Stage 
 (1) (2) (3) (4) (5) (6) (7) (8) 
Access to markets (nonagricultural individuals) 1.491*** (0.117) 1.454*** (0.114) 1.298*** (0.108) 1.277*** (0.134) 1.325*** (0.102) 0.262*** (0.082)  0.340** (0.154) 
Access to HP in Hydraulic Motors 1911       0.302*** (0.037)  
Mechanisms 
 Labor per hectare (family members)      0.259*** (0.056) 0.121*** (0.019) 0.247*** (0.060) 
 Capital per hectare      0.529*** (0.043) 0.018 (0.021) 0.526*** (0.043) 
Agrarian institutions 
 Owner-operators −0.730*** (0.273) −0.716*** (0.269) −0.528* (0.306) 0.181 (0.352) −0.842*** (0.270) −0.378** (0.165) −0.080 (0.078) −0.374** (0.165) 
 Rented tenants 0.163 (0.327) 0.172 (0.320) −0.286 (0.371) 0.312 (0.379) −0.186 (0.321) −0.080 (0.209) 0.397*** (0.122) −0.107 (0.214) 
 Sharecroppers 0.220 (0.207) 0.230 (0.204) 0.143 (0.226) 0.262 (0.248) −0.029 (0.198) −0.165 (0.136) −0.262*** (0.067) −0.145 (0.142) 
 Gini of farms (size) −1.358*** (0.336) −1.324*** (0.333) −0.676** (0.342) −1.754*** (0.504) −0.864** (0.338) −0.163 (0.202) −0.071 (0.107) −0.155 (0.202) 
 Gini of private ownerships (value) −2.799*** (0.565) −2.785*** (0.560) −1.350** (0.628) −2.867*** (0.700) −1.206** (0.549) 0.229 (0.339) −0.085 (0.183) 0.208 (0.342) 
 Average rent per ownership −0.128*** (0.032) −0.128*** (0.032) 0.017 (0.030) 0.465*** (0.039) 0.092*** (0.027) 0.116*** (0.020) 0.042*** (0.008) 0.112*** (0.021) 
 Share of land of collective entities −0.933*** (0.165) −0.938*** (0.162) −0.842*** (0.185) −1.703*** (0.192) −1.114*** (0.190) −0.426*** (0.118) 0.053 (0.049) −0.436*** (0.119) 
Socio-demographic variables 
 Literacy rate −1.037** (0.520) −1.024** (0.509) 0.433 (0.608) 0.935* (0.559) −0.340 (0.473) −0.304 (0.288) 0.313** (0.151) −0.355 (0.303) 
 Female literacy rate/male literacy rate −0.263 (0.601) −0.229 (0.586) 0.080 (0.663) −0.851 (0.642) −0.222 (0.554) −0.205 (0.351) 0.558*** (0.160) −0.251 (0.355) 
 Female–male ratio 0.816** (0.359) 0.829** (0.354) 0.037 (0.488) 0.503 (0.489) 0.633 (0.399) 0.398* (0.208) 0.229*** (0.084) 0.382* (0.207) 
 Share of inhabitants > 10 years old −0.640 (1.034) −0.626 (1.011) 1.247 (1.205) 1.311 (1.480) 1.336 (1.157) 0.839 (0.550) −0.197 (0.315) 0.824 (0.551) 
 Size of agricultural families −0.752*** (0.226) 0.517** (0.224) −0.239 (0.228) −1.353*** (0.260) −0.434** (0.219) −0.441*** (0.144) −0.121 (0.074) −0.436*** (0.146) 
 Share of dispersed population −0.251 (0.158) −0.255* (0.155) 0.032 (0.163) −0.682*** (0.181) −0.013 (0.153) 0.036 (0.093) 0.070* (0.042) 0.029 (0.093) 
Physical variables 
 Altitude −0.111*** (0.030) −0.115*** (0.029) −0.067** (0.033) −0.202*** (0.037) −0.115*** (0.030) −0.050** (0.022) −0.014 (0.010) −0.050** (0.022) 
 Terrain ruggedness 0.022 (0.026) 0.024 (0.025) −0.027 (0.028) −0.040 (0.029) 0.045* (0.025) 0.054*** (0.017) 0.013 (0.008) 0.053*** (0.017) 
 Share of malaria endemic area −0.174*** (0.065) −0.174*** (0.064) −0.001 (0.065) −0.173** (0.072) −0.097 (0.064) −0.052 (0.043) 0.014 (0.019) −0.053 (0.042) 
 Distance to the sea −0.042*** (0.013) −0.039*** (0.013) −0.074*** (0.014) −0.029** (0.014) −0.053*** (0.013) −0.004 (0.009) 0.005 (0.004) −0.005 (0.009) 
 Island dummy 0.813*** (0.226) 0.816*** (0.222) 0.802** (0.353) 0.378 (0.287) 1.069*** (0.304) 0.433** (0.194) −0.121* (0.073) 0.444** (0.194) 
 Latitude (km from Equator) −2.485* (1.389) −2.493* (1.362) −0.108 (1.771) −3.134* (1.655) −4.160*** (1.464) −3.457*** (0.931) −0.714 (0.473) −3.523*** (0.939) 
Rainfall regime 
 Total rainfall 1.116*** (0.399) 1.044*** (0.395) 0.587 (0.559) 1.641*** (0.524) 0.334 (0.518) −0.247 (0.269) −0.700*** (0.123) −0.156 (0.319) 
 CV total rainfall −0.198 (0.422) −0.192 (0.411) 0.642 (0.481) −0.086 (0.506) 0.587 (0.460) 0.297 (0.293) 0.006 (0.131) 0.297 (0.292) 
 Winter rainfall −0.151 (0.150) −0.137 (0.149) −0.117 (0.225) −0.790*** (0.202) −0.351* (0.193) −0.253*** (0.096) 0.186*** (0.051) −0.263*** (0.098) 
 Spring rainfall −0.647*** (0.212) −0.634*** (0.209) −0.344 (0.242) −0.760*** (0.234) −0.279 (0.235) 0.067 (0.120) 0.160*** (0.057) 0.034 (0.135) 
 Summer rainfall −0.068 (0.105) −0.053 (0.103) −0.105 (0.157) 0.019 (0.147) 0.134 (0.134) 0.204** (0.081) 0.163*** (0.035) 0.191** (0.085) 
 Autumn rainfall 0.304 (0.265) 0.319 (0.261) 0.147 (0.317) 0.403 (0.333) 0.374 (0.296) 0.213 (0.176) 0.314*** (0.085) 0.170 (0.197) 
 CV winter rainfall −0.312 (0.197) −0.280 (0.194) −0.040 (0.229) 0.304 (0.221) 0.062 (0.211) 0.156 (0.143) −0.044 (0.058) 0.150 (0.143) 
 CV spring rainfall −0.218 (0.267) −0.226 (0.262) −0.597** (0.300) 0.099 (0.292) −0.404 (0.283) −0.029 (0.177) −0.026 (0.081) −0.030 (0.177) 
 CV summer rainfall 0.123 (0.158) 0.135 (0.154) 0.152 (0.174) −0.092 (0.181) −0.088 (0.156) −0.204** (0.103) 0.018 (0.049) −0.209** (0.104) 
 CV autumn rainfall 0.165 (0.224) 0.170 (0.220) 0.379 (0.278) 0.033 (0.267) 0.286 (0.259) 0.042 (0.169) 0.039 (0.071) 0.055 (0.172) 
 Winter rain intensity 0.334* (0.187) 0.354* (0.185) 0.228 (0.244) 0.589** (0.231) 0.401* (0.211) 0.189 (0.120) 0.139** (0.060) 0.174 (0.122) 
 Spring rain intensity −0.308 (0.198) −0.317 (0.195) −0.129 (0.225) −0.077 (0.241) −0.142 (0.221) 0.008 (0.138) −0.123** (0.061) 0.020 (0.140) 
 Summer rain intensity −0.088 (0.115) −0.080 (0.113) −0.130 (0.146) −0.104 (0.142) −0.248* (0.132) −0.158* (0.095) 0.050 (0.035) −0.167* (0.094) 
 Autumn rain intensity −0.300 (0.196) −0.305 (0.193) −0.156 (0.237) −0.732** (0.230) −0.173 (0.217) −0.011 (0.132) −0.059 (0.056) −0.006 (0.133) 
 North dummy 0.044 (0.092) 0.061 (0.093) 0.022 (0.093) −0.096 (0.131) −0.011 (0.095) −0.038 (0.053) 0.004 (0.027) −0.036 (0.053) 
 South dummy −0.120 (0.079) −0.113 (0.079) 0.011 (0.086) 0.078 (0.100) −0.199** (0.084) −0.176*** (0.051) 0.028 (0.023) −0.175*** (0.051) 
 Constant 4.833 (11.815) 2.478 (11.571) −5.998 (15.156) 15.831 (13.958) 27.618** (12.439) 30.146*** (7.717) 15.492*** (3.838) 29.847*** (7.774) 
Number of obs. 732 732 732 732 732 732 732 732 
F-Statistic 27.68 36.32 27.63 105.26 41.02 142.07 67.72 139.89 
R-squared 0.613 0.646 0.568 0.836 0.664 0.870 0.778 0.870 
(All variables spatially filtered)
 
Dependent variable Agricultural labor per hectare (families) Agricultural labor per hectare (family members) Agricultural capital per hectare Rent per hectare Agricultural output per hectare
 
 OLS OLS OLS OLS OLS OLS IV IV 
       1st Stage 2nd Stage 
 (1) (2) (3) (4) (5) (6) (7) (8) 
Access to markets (nonagricultural individuals) 1.491*** (0.117) 1.454*** (0.114) 1.298*** (0.108) 1.277*** (0.134) 1.325*** (0.102) 0.262*** (0.082)  0.340** (0.154) 
Access to HP in Hydraulic Motors 1911       0.302*** (0.037)  
Mechanisms 
 Labor per hectare (family members)      0.259*** (0.056) 0.121*** (0.019) 0.247*** (0.060) 
 Capital per hectare      0.529*** (0.043) 0.018 (0.021) 0.526*** (0.043) 
Agrarian institutions 
 Owner-operators −0.730*** (0.273) −0.716*** (0.269) −0.528* (0.306) 0.181 (0.352) −0.842*** (0.270) −0.378** (0.165) −0.080 (0.078) −0.374** (0.165) 
 Rented tenants 0.163 (0.327) 0.172 (0.320) −0.286 (0.371) 0.312 (0.379) −0.186 (0.321) −0.080 (0.209) 0.397*** (0.122) −0.107 (0.214) 
 Sharecroppers 0.220 (0.207) 0.230 (0.204) 0.143 (0.226) 0.262 (0.248) −0.029 (0.198) −0.165 (0.136) −0.262*** (0.067) −0.145 (0.142) 
 Gini of farms (size) −1.358*** (0.336) −1.324*** (0.333) −0.676** (0.342) −1.754*** (0.504) −0.864** (0.338) −0.163 (0.202) −0.071 (0.107) −0.155 (0.202) 
 Gini of private ownerships (value) −2.799*** (0.565) −2.785*** (0.560) −1.350** (0.628) −2.867*** (0.700) −1.206** (0.549) 0.229 (0.339) −0.085 (0.183) 0.208 (0.342) 
 Average rent per ownership −0.128*** (0.032) −0.128*** (0.032) 0.017 (0.030) 0.465*** (0.039) 0.092*** (0.027) 0.116*** (0.020) 0.042*** (0.008) 0.112*** (0.021) 
 Share of land of collective entities −0.933*** (0.165) −0.938*** (0.162) −0.842*** (0.185) −1.703*** (0.192) −1.114*** (0.190) −0.426*** (0.118) 0.053 (0.049) −0.436*** (0.119) 
Socio-demographic variables 
 Literacy rate −1.037** (0.520) −1.024** (0.509) 0.433 (0.608) 0.935* (0.559) −0.340 (0.473) −0.304 (0.288) 0.313** (0.151) −0.355 (0.303) 
 Female literacy rate/male literacy rate −0.263 (0.601) −0.229 (0.586) 0.080 (0.663) −0.851 (0.642) −0.222 (0.554) −0.205 (0.351) 0.558*** (0.160) −0.251 (0.355) 
 Female–male ratio 0.816** (0.359) 0.829** (0.354) 0.037 (0.488) 0.503 (0.489) 0.633 (0.399) 0.398* (0.208) 0.229*** (0.084) 0.382* (0.207) 
 Share of inhabitants > 10 years old −0.640 (1.034) −0.626 (1.011) 1.247 (1.205) 1.311 (1.480) 1.336 (1.157) 0.839 (0.550) −0.197 (0.315) 0.824 (0.551) 
 Size of agricultural families −0.752*** (0.226) 0.517** (0.224) −0.239 (0.228) −1.353*** (0.260) −0.434** (0.219) −0.441*** (0.144) −0.121 (0.074) −0.436*** (0.146) 
 Share of dispersed population −0.251 (0.158) −0.255* (0.155) 0.032 (0.163) −0.682*** (0.181) −0.013 (0.153) 0.036 (0.093) 0.070* (0.042) 0.029 (0.093) 
Physical variables 
 Altitude −0.111*** (0.030) −0.115*** (0.029) −0.067** (0.033) −0.202*** (0.037) −0.115*** (0.030) −0.050** (0.022) −0.014 (0.010) −0.050** (0.022) 
 Terrain ruggedness 0.022 (0.026) 0.024 (0.025) −0.027 (0.028) −0.040 (0.029) 0.045* (0.025) 0.054*** (0.017) 0.013 (0.008) 0.053*** (0.017) 
 Share of malaria endemic area −0.174*** (0.065) −0.174*** (0.064) −0.001 (0.065) −0.173** (0.072) −0.097 (0.064) −0.052 (0.043) 0.014 (0.019) −0.053 (0.042) 
 Distance to the sea −0.042*** (0.013) −0.039*** (0.013) −0.074*** (0.014) −0.029** (0.014) −0.053*** (0.013) −0.004 (0.009) 0.005 (0.004) −0.005 (0.009) 
 Island dummy 0.813*** (0.226) 0.816*** (0.222) 0.802** (0.353) 0.378 (0.287) 1.069*** (0.304) 0.433** (0.194) −0.121* (0.073) 0.444** (0.194) 
 Latitude (km from Equator) −2.485* (1.389) −2.493* (1.362) −0.108 (1.771) −3.134* (1.655) −4.160*** (1.464) −3.457*** (0.931) −0.714 (0.473) −3.523*** (0.939) 
Rainfall regime 
 Total rainfall 1.116*** (0.399) 1.044*** (0.395) 0.587 (0.559) 1.641*** (0.524) 0.334 (0.518) −0.247 (0.269) −0.700*** (0.123) −0.156 (0.319) 
 CV total rainfall −0.198 (0.422) −0.192 (0.411) 0.642 (0.481) −0.086 (0.506) 0.587 (0.460) 0.297 (0.293) 0.006 (0.131) 0.297 (0.292) 
 Winter rainfall −0.151 (0.150) −0.137 (0.149) −0.117 (0.225) −0.790*** (0.202) −0.351* (0.193) −0.253*** (0.096) 0.186*** (0.051) −0.263*** (0.098) 
 Spring rainfall −0.647*** (0.212) −0.634*** (0.209) −0.344 (0.242) −0.760*** (0.234) −0.279 (0.235) 0.067 (0.120) 0.160*** (0.057) 0.034 (0.135) 
 Summer rainfall −0.068 (0.105) −0.053 (0.103) −0.105 (0.157) 0.019 (0.147) 0.134 (0.134) 0.204** (0.081) 0.163*** (0.035) 0.191** (0.085) 
 Autumn rainfall 0.304 (0.265) 0.319 (0.261) 0.147 (0.317) 0.403 (0.333) 0.374 (0.296) 0.213 (0.176) 0.314*** (0.085) 0.170 (0.197) 
 CV winter rainfall −0.312 (0.197) −0.280 (0.194) −0.040 (0.229) 0.304 (0.221) 0.062 (0.211) 0.156 (0.143) −0.044 (0.058) 0.150 (0.143) 
 CV spring rainfall −0.218 (0.267) −0.226 (0.262) −0.597** (0.300) 0.099 (0.292) −0.404 (0.283) −0.029 (0.177) −0.026 (0.081) −0.030 (0.177) 
 CV summer rainfall 0.123 (0.158) 0.135 (0.154) 0.152 (0.174) −0.092 (0.181) −0.088 (0.156) −0.204** (0.103) 0.018 (0.049) −0.209** (0.104) 
 CV autumn rainfall 0.165 (0.224) 0.170 (0.220) 0.379 (0.278) 0.033 (0.267) 0.286 (0.259) 0.042 (0.169) 0.039 (0.071) 0.055 (0.172) 
 Winter rain intensity 0.334* (0.187) 0.354* (0.185) 0.228 (0.244) 0.589** (0.231) 0.401* (0.211) 0.189 (0.120) 0.139** (0.060) 0.174 (0.122) 
 Spring rain intensity −0.308 (0.198) −0.317 (0.195) −0.129 (0.225) −0.077 (0.241) −0.142 (0.221) 0.008 (0.138) −0.123** (0.061) 0.020 (0.140) 
 Summer rain intensity −0.088 (0.115) −0.080 (0.113) −0.130 (0.146) −0.104 (0.142) −0.248* (0.132) −0.158* (0.095) 0.050 (0.035) −0.167* (0.094) 
 Autumn rain intensity −0.300 (0.196) −0.305 (0.193) −0.156 (0.237) −0.732** (0.230) −0.173 (0.217) −0.011 (0.132) −0.059 (0.056) −0.006 (0.133) 
 North dummy 0.044 (0.092) 0.061 (0.093) 0.022 (0.093) −0.096 (0.131) −0.011 (0.095) −0.038 (0.053) 0.004 (0.027) −0.036 (0.053) 
 South dummy −0.120 (0.079) −0.113 (0.079) 0.011 (0.086) 0.078 (0.100) −0.199** (0.084) −0.176*** (0.051) 0.028 (0.023) −0.175*** (0.051) 
 Constant 4.833 (11.815) 2.478 (11.571) −5.998 (15.156) 15.831 (13.958) 27.618** (12.439) 30.146*** (7.717) 15.492*** (3.838) 29.847*** (7.774) 
Number of obs. 732 732 732 732 732 732 732 732 
F-Statistic 27.68 36.32 27.63 105.26 41.02 142.07 67.72 139.89 
R-squared 0.613 0.646 0.568 0.836 0.664 0.870 0.778 0.870 

Notes: Robust standard errors in parenthesis; *significant at 10 percent, **significant at 5 percent, and ***significant at 1 percent. All variables in logarithms, when necessary transformed as ln(1 + x), and transformed applying the spatial filtering methodology suggested in Badinger et al. (2004).

Conclusion

In this paper I have explored a simple demand-led mechanism for an agricultural divergence in the spirit of the Von Thünen models of land use. The case study is Italy in the interwar years, a period during which agricultural output became increasingly concentrated in the Northern section of the country. Southern Italy, still performing well in the agricultural sector at the end of the nineteenth century, was unable to maintain its initial leadership during the first half of the twentieth century. This second divergence combined with a previous industrial one and produced the well-known and everlasting income gap between the North and the South.

With a detailed cross sectional database on agricultural output and capital, I test a model for the agricultural divergence connecting output per hectare with the access to markets via factor accumulation. The results strongly confirm the predictions of the model. Italian agriculture followed by and large a Von Thünen pattern around 1930. According to such a pattern, the quest for institutional failures in Southern agriculture is somewhat downplayed. Closer access to growing demand sources allowed higher levels of agricultural output by simply making the use of more inputs profitable.

The coefficients estimated in this paper suggest that, roughly speaking, doubling the access to markets would result in a doubling of output. These figures tally well with the evolution of the Italian economy during the period. Agricultural output in Northern Italy increased by 75 percent between 1891 and 1951, while in the South it did so only by a mere 25 percent (Federico 2003a, b, 2007). Because employment in agriculture did not change to a great degree during the entire period or across the country, this growing gap translated into diverging agrarian incomes. Between 1881 and 1951 the employment in sectors other than agriculture grew in the North from less than 3 to more than 6 million people (i.e., more than doubling), while it increased in the South from 2 to just 3 million people. In 1891 the average Northern peasant had to feed 0.6 individuals employed outside the sector, as did the average Southern one. Sixty years later, the same ratio had boomed to 1.7 in the North, while it was languishing at a level of 0.8 in the South.

The model explored here fully explains the agricultural divergence and connects it with the industrial divergence. The key difference between Northern and Southern agriculture during the period laid outside the sector: it was the additional three million people employed in Northern industry and in the urban services sector that pushed the demand for agricultural products much farther in the North than in the South. This demand push provided the incentives for additional investment in land reclamation, machinery, irrigation works, tree planting, and livestock. It may also have provided incentives for further efficiency gains. Southern agriculture simply did not have such a set of incentives and evolved accordingly.

Nonetheless, the insights gained from this analysis suggest the need for further research. Further quantitative research on Italian agriculture during this period is surely desirable, particularly in terms of the estimation of the regional or provincial time series of the agricultural output (and possibly capital). While the model is presented here in terms of a single homogeneous product, a more sophisticated Von Thünen formulation will predict a succession of crops in a featureless plain (determined by product prices and by technical coefficients of production). With some additional research, this side of the model (as well as possible deviations from it) could be explored.

There has also been recent focus on the role played by the human and social capital in the emergence of the Italian regional divide (Felice 2010, 2012). While the analytical framework of this paper does not need to factor in such determinants, these explanations may be viewed as complementary rather than as alternatives. After all, human and social capital are factors of production and a closer access to markets can yield higher returns which, in turn, stimulate further accumulation. Hence, the interaction between access to markets and the “intangible” factors of production remains a promising field of research.

During this period international trade collapsed and hence the importance of access to the domestic market considerably increased. Assessing the relative impact of the drop in external trade on Northern and Southern agriculture remains part of the research agenda. Southern agriculture produced export products such as oil, citrus, and wine, and the collapse of the international outlays for these products may have added a nonnegligible shock to the Southern economy, which was already disadvantaged by its relatively peripheral position in terms of the access to domestic markets. The accumulation of capital, as well as internal migrations, may have been shaped at least to some degree by this shock, and this may have had long lasting effects on the Southern economy.

Additional research on each of these subjects would highlight another side of the advantage of Northern agriculture during this period. All these factors notwithstanding, this paper demonstrates that it is not surprising that the output and factors of production concentrated in the increasingly industrialized Northern regions. Indeed, the agricultural and the industrial Italian divergences were not two but one and the same, with the former being the consequence and the latter the cause. Interwar Italy happens to be the clearest example of Adam Smith's insight: “It is thus that through the greater part of Europe the commerce and manufactures of cities, instead of being the effect, have been the cause and occasion of the improvement and cultivation of the country9”.

Supplementary material

Supplementary material is available at EREH online.

Acknowledgements

I thank Giovanni Federico for his criticisms and suggestions, as well as participants at the Pisa FRESH Meeting 2012, at the EUI Economic History Workshop, at the UC3M Economic History Workshop and at the EHS Conference 2013 for their comments. Nikolaus Wolf and Juan Carmona made also useful comments on a previous version of the paper. The paper also benefited from discussions with Joan Rosés. Two anonymous referees, the editorial board of the Review and participants at the Fast Track Meeting in London helped in improving substantially the paper. Remaining errors and interpretations are solely my responsibility. Support is acknowledged from the Spanish Ministry of Science and Innovation project HAR2010-20684-C02-01.

References

A'hearn
B.
Venables
A.
Internal Geography and External Trade: Regional Disparities in Italy, 1861–2011
2011
Rome
Banca d'Italia
 
Economic History Working Papers
Badinger
H.
Müller
W.
Tondl
G.
Regional convergence in the European Union (1985–1999): a spatial dynamic panel analysis
Regional Studies
 , 
2004
, vol. 
38
 (pg. 
241
-
53
)
Beckmann
M.J.
Von Thünen revisted: a neoclassical land use model
Swedish Journal of Economics
 , 
1972
, vol. 
74
 (pg. 
1
-
7
)
Brunetti
A.
Felice
E.
Vecchi
G.
Vecchi
G.
Reddito
2011
(pg. 
209
-
34
2011
Carmona
J.
Rosés
J.
Land markets and agrarian backwardness (Spain, 1904–1934)
European Review of Economic History
 , 
2012
, vol. 
16
 (pg. 
74
-
96
)
Ciocca
P.
Toniolo
G.
L'economia italiana nel periodo fascista
 , 
1976
Bologna
Il Mulino
Cohen
J.S.
Ciocca
P.
Toniolo
G.
Rapporti agricoltura-industria e sviluppo agricolo
1976
(pg. 
379
-
407
1976
Cohen
J.S.
Federico
G.
The Growth of the Italian Economy, 1820–1960
 , 
2001
Cambridge
Cambridge University Press
Cohen
J.S.
Galassi
F.L.
Sharecropping and productivity: “feudal residues” in Italian agriculture
The Economic History Review
 , 
1990
, vol. 
4
 (pg. 
646
-
56
XLIII
Combes
P.-P.
Lafourcade
M.
Thisse
J.-F.
Toutain
J.-C.
The rise and fall of spatial inequalities in France: a long-run perspective
Explorations in Economic History
 , 
2011
, vol. 
48
 (pg. 
243
-
71
)
Daniele
V.
Malanima
P.
Il divario Nord-Sud in Italia, 1861-2011
 , 
2011
Soveria Mannelli
Rubbettino
Federico
G.
Rey
G.M.
Il valore aggiunto dell'agricoltura
1992. I conti economici dell'Italia 2. Una stima del valore aggiunto per il 1911
 , 
1992
Roma-Bari
Laterza
(pg. 
3
-
104
)
Federico
G.
Rey
G.M.
Una stima del valore aggiunto in agricoltura
2000 . I conti economici dell'Italia 3.2. Il valore aggiunto per il 1891, 1938 e 1951
 , 
2000
Roma-Bari
Laterza
(pg. 
3
-
112
)
Federico
G.
Ciocca
P.L.
Toniolo
G.
L'agricoltura italiana: successo o fallimento?
Storia economica d'Italia 3.1. Industrie, mercati e istituzioni
 , 
2003a
Roma-Bari
Laterza
(pg. 
71
-
98
)
Federico
G.
Le nuove stime della produzione agricola italiana, 1860-1910: primi risultati ed implicazioni
Rivista di Storia Economica
 , 
2003b
, vol. 
XIX
 (pg. 
359
-
81
, December
Federico
G.
Feeding the World. An Economic History of Agriculture, 1800–2000
 , 
2005
Princeton
Princeton University Press
Federico
G.
Ma l'agricoltura meridionale era davvero arretrata?
Rivista di Politica Economica
 , 
2007
, vol. 
III-IV
 (pg. 
1
-
24
March–April
Federico
G.
Lains
P.
Pinilla
V.
Agriculture and economi growth in Italy, 1870–1939
2009. Agriculture and Economic Development in Europe since 1870
 , 
2009
Oxon and New York
Routledge
(pg. 
234
-
54
)
Felice
E.
Divari regionali e intervento pubblico. Per una rilettura dello sviluppo in Italia
 , 
2007
Bologna
il Mulino
Felice
E.
Regional development: reviewing the Italian mosaic
Journal of Modern Italian Studies
 , 
2010
, vol. 
15
 (pg. 
64
-
80
)
Felice
E.
Regional value added in Italy, 1891–2001, and the foundation of a long-term picture
The Economic History Review
 , 
2011
, vol. 
3
 (pg. 
929
-
50
)
Felice
E.
Regional convergence in Italy, 1891–2001: testing human and social capital
Cliometrica
 , 
2012
, vol. 
6
 (pg. 
267
-
306
)
Getis
A.
Ord
J.K.
The analysis of spatial association by use of distance statistics
Geographical Analysis
 , 
1992
, vol. 
24
 (pg. 
189
-
206
)
INEA (Istituto Nazionale di Economia Agraria)
La distribuzione della proprietà fondiaria in Italia
 , 
1946–1948
Rome
 
13 regional volumes
ISTAT (Istituto Centrale di Statistica del Regno d'Italia)
VII Censimento Generale della Popolazione
 , 
1931–1935
Rome
 
21 Aprile 1931. Vol. III Fascicoli Provinciali, 95 provincial volumes
ISTAT (Istituto Centrale di Statistica del Regno d'Italia)
Catasto Agrario del Regno d'Italia 1929
 , 
1932–1937
Rome
 
95 provincial volumes
ISTAT (Istituto Centrale di Statistica del Regno d'Italia)
Censimento Generale dell'Agricoltura, Vol. I Censimento del Bestiame, Vol. II Censimento delle Aziende Agricole, Vol. III Censimento delle Bonifiche Idrauliche
 , 
1933–1935
Rome
ISTAT (Istituto Centrale di Statistica del Regno d'Italia)
Dizionario dei Comuni e delle Frazioni di Comune alla Data del 21 Aprile 1936
 , 
1938
Rome
ISTAT (Istituto Centrale di Statistica del Regno d'Italia)
VIII Censimento Generale della Popolazione
 , 
1939
Rome
 
21 Aprile 1936. Vol. IV Professioni
Iuzzolino
G.
Pellegrini
G.
Viesti
G.
Convergence among Italian regions, 1861–2011
2011
Rome
Banca d'Italia
 
Economic History Working Papers
Kim
S.
Expansion of markets and the geographic distribution of economic activities: the trends in U.S. regional manufacturing structure, 1860–1987
Quarterly Journal of Economics
 , 
1995
, vol. 
110
 (pg. 
881
-
908
)
Klein
A.
Crafts
N.
Making sense of the manufacturing belt: determinants of U.S. industrial location, 1880–1920
Journal of Economic Geography
 , 
2012
, vol. 
12
 (pg. 
775
-
807
)
Kopsidis
M.
Wolf
N.
Agricultural productivity across Prussia During the industrial revolution: a Thünen perspective
The Journal of Economic History
 , 
2012
, vol. 
72
 (pg. 
634
-
70
)
Krugman
P.
Increasing returns and economic geography
Journal of Political Economy
 , 
1991
, vol. 
99
 (pg. 
483
-
99
)
Martínez-Galarraga
J.
The determinants of industrial location in Spain, 1856–1929
Explorations in Economic History
 , 
2012
, vol. 
49
 (pg. 
255
-
75
)
Ministero dei Lavori Pubblici
1936–1939
 
Annali Idrologici. Published by the different Basin Authorities
Ministero di Agricoltura, Industria e Commercio
Censimento degli Opifici e delle Imprese Industriali al 10 Giugno 1911. Vol. IV, Dati analitici concernenti il numero, il personale e la forza motrice di tutte le imprese censite
 , 
1914
Rome
Puga
D.
The rise and fall of spatial inequalities
European Economic Review
 , 
1999
, vol. 
43
 (pg. 
303
-
34
)
Rosés
J.R.
Martínez-Galarraga
J.
Tirado
D.A.
The Upswing in Regional Income Inequality in Spain 1860–1930
Explorations in Economic History
 , 
2010
, vol. 
47
 (pg. 
244
-
57
)
Samuelson
P.
Thünen at two hundred
Journal of Economic Literature
 , 
1983
, vol. 
21
 (pg. 
1468
-
88
)
Sereni
E.
Il capitalismo nelle campagne (1860–1900)
 , 
1977
Torino
Einaudi
 
(or. ed. 1947)
Simpson
J.
Spanish Agriculture: the Long Siesta, 1765–1965
 , 
1995
Cambridge
Cambridge University Press
Statistisches Reichsamt
Statistisches Jahrbuch für das Deutsche Reich 1941–1942
 , 
1942
Berlin
Toniolo
G.
L'economia dell'Italia fascista
 , 
1980
Roma-Bari
Laterza
Vecchi
G.
In ricchezza e in povertà. Il benessere degli Italiani dall'Unità a oggi
 , 
2011
Bologna
Il Mulino
Vitali
O.
La popolazione attiva in agricoltura attraverso i censimenti italiani
 , 
1968
Rome
Istituto di Demografia
Vollrath
D.
Land distribution and international agricultural productivity
American Journal of Agricultural Economics
 , 
2007
, vol. 
89
 (pg. 
201
-
16
)
Von Thünen
J.H.
Die isolierte Staat in Beziehung auf Landwirtschaft und Nationalökonomie
 , 
1826
Jena
 
(1910 ed.)
Wolf
N.
Endowments vs. market potential: what explains the relocation of industry after the Polish reunification?
Explorations in Economic History
 , 
2007
, vol. 
44
 (pg. 
22
-
42
)
Zamagni
V.
Dalla periferia al Centro. La seconda rinascita economica dell'Italia (1861–1990)
 , 
1993
Bologna
Il Mulino
1
Unfortunately, data are too scarce to allow for the quantification of each component of the divergence: there are only regional estimates of agricultural output for four benchmark years (Federico 2003a, b) and the regional time series of the agricultural capital are unavailable at all.
2
It is a paper produced in the context of a research project promoted by the Bank of Italy in order to celebrate the 150th anniversary of the unification of (the main part of) the country. It is desirable that more research along the lines depicted in A'Hearn and Venables (2011) will appear, especially because the main contribution of the paper is to set the main framework and a general interpretation.
3
Italy was still on the way to homogenizing its cadastral system more than 70 years after the unification of the country. The simultaneous revision of the cadastral estimates, which was aimed at raising taxes in order to fund the coming war effort, began in 1939 and in 1943 was already finished. The criteria that informed the rent estimates were aimed at capturing the economic rent as the marginal product of land and thus were, from an agronomic point of view, correct for the first time in Italian history; indeed, they are still in force.
4
First, the definition of “urban” may be controversial. Second, the demand for agricultural products could come from proper cities as well as from a large set of legally small municipalities with high population density and high levels of industrialization (a definition matching the features of some industrial districts in Northern Italy). Third, the rural population engaged in agriculture could be considered both a supplier and a demander of agricultural products, complicating the analysis somewhat.
5
The gravity center is defined as the weighted sum of coordinates (latitude and longitude) of all the municipalities included in the agrarian zone, with the population in 1931 being the weights. The data on latitude and longitude for every Italian municipality at the 1931 boundaries were provided by the Italian Army's Geographical Institute and published in the 95 volumes of the 1929 Agrarian Cadastre. The coordinates have been converted from degrees into kilometers in order to take into account the curvature of the Earth, as described in the Supplementary material, Appendix.
6
The data were taken from the Population Census of 1936 (ISTAT 1939), because the 1931 Population Census did not report employment data at the agrarian zone level, and the respective shares of every category are not likely to have changed greatly in 5 years.
7
Other estimates using alternative definitions yield basically the same results, and are available upon request.
8
Columns 1–4 can be compared with Table 4, columns 5 and 6 with Table 5 and columns 7 and 8 with columns 7 and 8 in Table 6. Including the access to foreign markets in the regression produces results similar to those reported in Table 7, with a similar further collapse of the statistical significance of the coefficient of the access to markets.
9
Cited in Simpson (1995), p.177.

Supplementary data