Origins of the Sicilian Ma�a:The Market for Lemons

Arcangelo Dimico� Alessia Isopi Ola Olsson

June 23, 2014

Abstract

In this paper, we study the emergence of an extractive institution that hamperedeconomic development in Italy for more than a century: the Sicilian ma�a. Since its�rst appearance in the late 1800s, the origins of the Sicilian ma�a have remained apuzzle. In this paper, we develop the argument that the ma�a arose as a response toan exogenous shock in the demand for oranges and lemons, following Lind�s discoveryin the late 18th century that citrus fruits cured scurvy. More speci�cally, we outlinea market structure-hypothesis, contending that ma�a emerged in certain towns where�rms made very high pro�ts from citrus production due to imperfect competition. Thesector was characterized by substantial �xed costs that acted as an important barrierto entry. The ma�a arose out of the need to protect citrus production from predation.Using the original data from a parliamentary inquiry in 1881-86 on Sicilian towns, theDamiani Inquiry, we show that ma�a presence is strongly related to the production oforanges and lemons. The results hold when di¤erent data sources and di¤erent controlsare employed. Our �ndings contrast recent works that emphasize the role played by landreforms and by sulphur mining.

Keywords: ma�a, Sicily, protection, barriers to entry, dominant positionJEL Codes: K42, P48, L1

1 Introduction

The Sicilian ma�a is arguably one of the most infamous institutions in the Western world.

After its �rst appearance in Sicily in the 1870s, it soon in�ltrated the economic and political

spheres of Italy and the United States and has, at times, been considered a serious threat to

the rule of law in both countries.1 Although outcomes of the ma�a�s actions such as murders,

�Corresponding author. Email: a.dimico@qub.ac.uk. Special thanks to Giuliano Isopi for excellent researchassistance. We gratefully acknowledge comments from Gani Aldashev, Jean-Marie Baland, Michael Bleaney,Paolo Casini, Chris Colvin, Giacomo de Luca, Stefano Fenoaltea, Halvor Mehlum, Kalle Moene, KyriacosNeanidis, Jean-Philippe Platteau, Oleg Shchetinin, Yves Zenou, and seminar participants in Nottingham,Brown University, Gothenburg, Leuven, Birkbeck College, the EEA-ESEM meetings in Malaga, the RESConference in Manchester, Namur, Oslo, Stockholm and York.

1The �rst evidence of the presence of a secret sect (referred to as cosca) we are aware of is an account from1872 by Dr Galati, a landlord who used to own a lemon grove just outside Palermo, who wrote about a �manof honor� who made increasing use of violence and extortion to force him to sell his lemon grove (Dickie,2004). When the Minister of Home A¤airs learned about Galati�s note, he soon asked for a written reportfrom the chief of police in Palermo and then, ordered two parliamentary inquiries (the Bonfadini Inquiry in1876 and later the Damiani Inquiry in 1881-5) to better understand the economic conditions and the level ofcrime in Sicily.

1

bombings, and embezzlement of public money have been readily observed during the last

140 years, its origins have largely remained a mystery.

In this paper, we provide a study of the origins of the Sicilian ma�a using a unique

dataset from the end of the 19th century. The main hypothesis that we investigate is that

the origins of Sicilian ma�a are strongly associated with an exogenous demand shock for

lemons around 1800, following James Lind�s discovery that citrus fruits cured scurvy. Given

Sicily�s dominant position in the international market, the demand increase led to a very large

in�ow of revenues to citrus-producing towns in Sicily by 1850. We argue that the primary

source of these very high pro�ts to certain towns is to be found in a particular market

imperfection: the large and geographically varying �xed costs of production that acted as a

barrier to entry in the market for lemons. Because of very speci�c requirements, citrus trees

could only be cultivated in certain areas, guaranteeing substantial pro�ts to only few local

producers.2 The combination of high pro�ts, a weak rule of law, a low level of interpersonal

trust, and a high level of local poverty made lemon producers a target for predation. The

newly formed government after Italian independence in 1861 did not have the strength and

the means for e¤ectively enforcing private property rights and lemon producers, therefore,

resorted to hiring ma�a a¢ liates for private protection.

On the basis of a formal model of citrus market imperfections featuring households,

producers, and a ma�a organization, we test the main implications of the theory using a

dataset from Sicilian towns and districts gathered from a parliamentary inquiry conducted

between 1881-86 (Damiani, 1886). Our results indicate that ma�a presence in the 1880s is

strongly associated with the prevalence of citrus cultivation, after controlling for a number of

other potential covariates. No other crop or industry has a robust impact on ma�a activity.

We believe these �ndings to be consistent with a market structure-explanation for the origins

of ma�a. The results continue to hold when we include several control variables, address a

possible endogeneity issue using data on the climatic conditions, and adopt two alternative

dependent variables collected and coded from a later source. We conclude by showing that

the historical distribution of ma�a is a strong predictor for its more recent activity.

Our paper relates to several di¤erent strands of literature.3 First, it is related to the

literature on the historical emergence of an �extractive�institution that hampers economic

development and that can appear at critical junctures in a country�s history (Acemoglu

et al., 2006; Acemoglu and Robinson, 2012). The ma�a is undoubtedly an example of this,

emerging during a critical period in the Italian history (i.e. Italian uni�cation). Our analysis

though departs from this strand, since we emphasize the economic ormarket structure-related

factors behind ma�a organization rather than its political origins (such as the role played

by a weak and oppressive Bourbon state in Sicily with substantial social inequalities, as

discussed further below).

Our results are also strongly associated with research on the �curse of natural resources�

2The high �xed cost were the result of costs of planting trees and waiting several years for them to grow,the need to build protective walls to keep thieves out, construction of irrigation, etc. Due to a large regionalvariation in the climate and soil suitability for growing lemons, the �xed costs of starting up a cultivationwere very di¤erent across towns.

3Please see the working paper version for a more extensive review.

2

(see Van der Ploeg (2011) for a recent overview). We claim that the economic boom in

international citrus demand, and the subsequent rise of Sicilian exports in the second half

of the 19th century, is a key factor behind the rise of ma�a. This is also consistent with

the more recent �nding that windfall gains from natural resources are often associated with

intense rent seeking and patronage politics. For instance, Sala-i-Martin and Subramanian

(2003) argue that political corruption related to oil revenues hampered Nigeria�s growth for

decades. Acemoglu et al. (2004) show how mineral wealth in Zaire allowed President Mobutu

to buy o¤ political challengers. A recurrent theme in this tradition is that resource windfalls

might actually destabilize and deteriorate institutions, if key groups in the society believe

that predation is more pro�table than production (Mehlum et al., 2006; Congdon Fors and

Olsson, 2007).

Another literature that our analysis connects with is the one on the economic analysis of

organized crime, which focuses on weak institutions, predation, and enforcement of property

rights (Fiorentini, 1999; Grossman, 1995; Anderson, 1995; Skaperdas and Syropoulus, 1995;

Skaperdas, 2001). Grossman (1995) and Skaperdas (2001) both consider ma�a as an alterna-

tive enforcer of property rights. Using a model with two actors (a self-governing community

and ma�a) and potential robbers, Skaperdas (2001) shows that in the absence of an enforcer

of property rights, ma�a can represent a sort of second best solution.4 Regarding the eco-

nomic costs of organized crime, Reuter (1987) and Gambetta and Reuter (1995) analyze the

e¤ect of organized crime on the enforcement of cartel agreements in legal markets.

Related to the weak institutions-hypothesis, there are two other factors that may deter-

mine the development of a private provider of protection: a lack of both social capital and

public trust. Putnam (1993), for example, analyzes the loss of social capital in Southern Italy

due to the repeated foreign dominations experienced in the region. Gambetta (1996) con-

siders instead private trust rather than public trust as very important for the development

of ma�a in southern Italy. Social capital together with a loss of public trust may further

a¤ect the development of organized crime because of kinship relations, corruption, etc. (see

Fukuyama 2000; Gambetta, 2009; Levi, 1996; Hardin, 1999; Newton, 2001).

Yet, our paper is most closely related to Bandiera (2003). Bandiera�s main hypothesis is

that the increase in land fragmentation, following the Bourbon-era land reforms (1816-1860),

provided the breeding ground for ma�a protection: a higher number of land owners increase

the need for private protection. In Bandiera�s model, a key feature is that protection of one

producer generates a negative externality on other producers, since it makes them more likely

to become objects of predation. In an empirical section where she uses information from the

Summary Report presented to the Italian parliament by Damiani (1886), Bandiera (2003)

�nds that a variable capturing di¤erences in land fragmentation is a signi�cant determinant

of ma�a presence.5

4The idea of a weak state and private protection is well illustrated by Don Calo Vizzini, one of the historicalbosses of the ma�a in Villalba. In an interview with Indro Montanelli, he said that �...the fact is that inevery society there has to be a category of people who straighten things out when situations get complicated.Usually they are functionaries of the state. Where the state is not present, or where it does not have su¢ cientstrenght, this is done by private individuals� (Montanelli, 1949).

5The information available in the Damiani Inquiry (see Footnote 1) has previously been used also by other

3

While our analysis also identi�es landowners�demand for private protection as the main

process through which the ma�a was mobilized, we explicitly focus on the role of the market

structure rather than on land fragmentation. With respect to Bandiera (2003), we improve

the size of the sample by using the original Damiani survey (1883) where pretori (lower court

judges) provided answers on the causes of crime (see Figures A2 and A3 in the Appendix).

The original copy of the Damiani Inquiry allows us to extend the analysis from the 70 towns

located in the western part of the island (Bandiera (2003)) to almost all available Sicilian

towns (143 in total) for which pretori provided answers. With this more complete sample,

we �nd that the land fragmentation variable indeed explains some of the variation in ma�a

presence. However, we also �nd that the most robust determinant of ma�a activity is the

production of citrus fruits.

The paper by Buonanno et al. (2012) also studies the importance of export markets for

ma�a appearance using data from Cutrera (1900), a police o¢ cer in Palermo. Cutrera uses as

sources Colajanni (1900), Alongi (1887) and other data from local police o¢ ces to elaborate

a map of Sicily where the intensity of ma�a activity is outlined for every city. 6Even though

the data provide �gures on the level of ma�a for most of the Sicilian cities at the beginning

of the 20th century, they refer to a period of almost twenty years later than the Inquiry. In

those years, ma�a extended its activity to cities that initially were not a¤ected and hence,

we believe that data from Cutrera are more appropriate for understanding the evolution of

the ma�a phenomenon over time.7Buonanno et al. (2012) �nd that sulphur production had

a strong association with ma�a presence in 1900. However, they do not develop an explicit

argument for why export revenues were associated with ma�a revenues in certain sectors.

Furthermore, our main result that citrus production explained the presence of ma�a holds

even when we use Cutrera�s data.

Besides the economic literature, our analysis is also related to a long tradition of works in

anthropology, sociology and history on the Sicilian ma�a. The classical contributions include

early investigations from Villari (1875), Sonnino and Franchetti (1877), and Colajanni (1885,

1895). In recent years, the origin of the ma�a has also been discussed in Gambetta (1996),

Dickie (2004), and Lupo (2009).8 While Lupo (2009) and Dickie (2004) consider pro�ts

related to the lemon industry in the Western part of the island as a pre-condition for the

development of the ma�a, Gambetta (1996) focuses on the division of land resulting from

the abolition of feudalism and other policies introduced by the Italian government after 1860

(i.e., the sale of land owned by the church and the crown before the uni�cation). These

policies opened a market for private protection, where the ma�a acted as an incumbent.

scholars studying the origins of the Sicilian ma�a. See for instance Colajanni (1885, 1895), Hess (1973),Arlacchi (1986), Catanzaro (1992), Gambetta (1996), Dickie (2004) and Lupo (2007).

6Alongi (1887) and Colajanni (1900) themselves use the information available from the original inquiry.Therefore, their books represent a further elaboration of the results collected by the Damiani Inquiry, whichcould potentially add a bias in the data provided in Cutrera (1900).

7This is supported by Gambetta (1996), who argues that in the period between the late 1870s and late1890s, the ma�a evolved quite markedly as a result of �economic and political con�icts among local groups,in connection with the institutional change that a¤ected Italy between 1869-1890�(Gambetta, 1996, pg.83).

8See Lupo (2009) for a general history and Monroe (1909) for a description of the agricultural practicesin Sicily at the time.

4

The extensive literature discussed above provides plausible explanations for the origins of

the ma�a. Yet, with the exception of Bandiera (2003) and Buonanno et al. (2012), it is still

di¢ cult to understand, on the basis of these works, why we observe a substantial variation

in ma�a activity across provinces experiencing very similar social, economic and political

conditions. If a weak state, a high regulatory burden, and a lack of public trust are the

factors that matter for the development of ma�a, then we should not observe any province

variation within the territory at hand. However, this is not the case as across counties and

villages exposed to these same conditions there is a notable di¤erence in ma�a presence:

organized forms of crime initially appeared only in a small number of localities and then

spread all over the region.

The combined resource boom- and market structure-hypothesis advanced in our paper

thus not only complements existing theories of ma�a emergence (for instance those focusing

on political factors) but is also consistent with the timing of the origins of ma�a. It also

allows us to explain the cross-regional variation across Sicily.

In summary, this paper contributes to the existing literature in the following ways: First,

it provides a formal model of how the market structure and the prevalence of cross-sectional

variation in �xed costs a¤ect the demand for ma�a. Second, it o¤ers the most comprehen-

sive empirical analysis to date on the origins and the persistence of ma�a since the 1880s,

providing evidence for a novel explanation.

The paper is structured as follows: Section 2 provides us with background information

on the Sicilian economy and the role of citrus fruits. Section 3 outlines the formal model.

Section 4 includes the econometric speci�cation and a discussion of the data, whereas the

main empirical results are found in Section 5. Section 6 addresses some robustness tests and

Section 7 shows results on ma�a persistence. The paper is ended by Section 8.

2 The Sicilian economy and the role of citrus fruits

Sicily is the largest island in the Mediterranean. It is easily accessible by the sea from all

directions and has a varied and fertile territory and, above all, an unparalleled location:

the island in fact has been both a gateway and a crossroads, on the one hand dividing the

eastern and western Mediterranean, on the other linking Europe and Africa as a stepping

stone (Finley et al., 1987).

Because of the island�s importance and strategic location, its past has been marked by

continuous foreign domination and the list of migrants and colonizers is long: First the

Greeks, then the Romans followed by the Byzantines, then the Arabs, the Normans, the

Spanish, the French, and then the Spanish again. This long period of di¤erent foreign rulers

had an important impact on the the historical development of Sicily. The material advantages

of the foreign powers were created at the expense of the Sicilians through rents, taxes, and

plain looting. This parasitical interest inevitably did great harm to the countryside and to

its people.

At the beginning of the 19th century, overseas commerce was mostly carried out by the

5

French and a native industry in Sicily hardly existed except at household level. Sicily lacked

the main prerequisites for an industrial revolution: no iron, no coal, no navigable canals and

rivers and very poor internal communications. The middle class remained professional and

bureaucratic rather than commercial and industrial, while the leaders of the society lived o¤

rents.

Before 1861, the vast majority of the population was employed in agriculture. The

production system in the agricultural sector re�ected the typical feudal system. Villari

(1875) reckons that there were three main social classes in Sicily: 1) landlords, 2) a middle

class (gabelloti), 3) and peasants who were normally exploited by the gabelloti. The gabelloti

leased the land from landlords and then rented small pieces of it to the peasants. Peasants

worked the land and gave back to the gabelloti a share of the harvest depending on the kind

of contract stipulated. The contracts were relatively short and generally designed to exploit

the peasants. When the yield was not enough, peasants usually had to borrow from the

gabelloti at interest rates that made it impossible to pay back the debt.

The situation did not change much after Sicily joined the Reign of Italy in 1861. In 1887

the share of citizens that owned the land was still the lowest in Italy with an average of less

than 2.05 owners per 100 citizens, compared to, for instance, 15 owners per 100 in Piedmont

(Colajanni, 1885).9 In addition, almost 56 percent of the population employed in agriculture

owned less than one hectare of land and most of them were hired by the landlord on a daily

basis at less than one lira per day rate.

Despite its underdeveloped economy, Sicily was a leading producer of wheat, olive oil,

wine, and citrus fruits. The international demand for lemons started to increase from the late

1700s when lemons and, in particular, lemon juice became a standard preventive treatment

against scurvy. Scienti�c support for the theory that consumption of citrus fruits cured

scurvy was established by James Lind, a British naval o¢ cer and surgeon, in the latter part

of the 18th century. In the words of Baron (2009): �The Sick and Hurt Commission agreed

to supply all naval ships on foreign service with lemon juice, extended in 1799 to all the ships

on the British coast. Between 1795 and 1814 the admiralty issued 1.6 million gallons of

lemon juice. Sweet lemons were imported, especially from the Mediterranean region turning

Sicily into a vast lemon juice factory �(Baron, 2009; pg. 324).

In the late nineteenth century, Sicilian production of citrus fruits represented almost 73

percent of the total production in Italy (Pescosolido, 2010).10 Table A1 in the Appendix

reports the distribution of lemon trees in southern Italy in 1898. Palermo and Messina were

the two provinces with the largest absolute number of lemon trees, accounting for almost

59.5 percent of the total trees in Sicily, followed by Catania with approximately 12 percent.

Outside Sicily, in the Calabria region, Reggio Calabria was also a large producer of lemons

with an absolute number of trees equal to 1,232,675 (almost 18 percent).11 The total surface

9 In 1840, 127 princes, 78 dukes, 130 marquises, and an unknown number of earls and barons had completecontrol over most of the land (Travelyan, 2001).10Other reasons why Sicily may have had a dominant position in this sector are historical (related to the fact

that the plant developed in the Mediterranean area since the Roman times) and geographical (the importanceof the Mediterranean area for international trade).11Some trees were also planted in the central part of Italy (almost 798,214 trees) and in north of Italy

6

area devoted to the citrus production went from 7,695 hectares in 1853 to 26,840 hectares in

1880 (Pescosolido, 2010). The expansion was a direct result of the large returns associated

with the boom in demand. Monroe (1909) estimates that revenues were almost $200 per acre

(in 1908 US dollars), providing a net pro�t of more than $150 per acre. The importance of

these �gures is further emphasized by Dickie (2004) when he argues that �citrus cultivation

yielded more than sixty times the average pro�t per hectare for the rest of the island� (Dickie,

2004, p.39).

In the period 1881-85, the quantity of citrus exported was almost 949,000 units, compared

to 250,000 units in 1850 (Pescosolido, 2010). A large share of this production was exported

to the US which was one of the largest markets for Sicilian citrus exports. A combination of

factors contributed to this outcome: a favorable international context, elimination of exports

duties, and a considerable improvement of transports.

Table 1 shows �gures on the lemon trade between Italy and the US in 1898-1903. The

left-hand side of the table shows the total Italian lemons exports and the relative percentage

exported to the US. The right-hand side shows the total US lemons imports and the estimated

percentage coming from Italy.12 The average quantity of lemons exported from Italy (and

therefore mainly from Sicily) over this period amounts to 389 million pounds and the average

share of fruits imported from the US is almost 34 percent of the total Italian production.

On the other hand, most lemons imported from the US in this period came from Italy.

Calculating the total Italian exports to the US, using percentages to the right in the table,

we can estimate that almost 78.4 percent of the total US lemons imports between 1898-

1903 came from Italy. Besides the US, the UK and Austria were also two large importers

of lemons. Over the decade 1898-1908, the UK imported between 17.7 to 25 percent of

the total Italian lemons exports, and the Austro-Hungary imported between 14.4 and 22.8

percent (Powell, 1909).

Table 1: Total Italian exports of lemon and total United States imports

Table 2 shows the total Italian exports of citrate of lime (i.e. a soft drink) as well as

its total imports into the USA during the 1899-1903 period. The US is, again, the most

important market for citrate together with France and the UK. According to Powell (1909)

the total quantity imported by these last two countries together was at least as large as the

one for the US. A similar pattern can also be shown for the total exports of essential oils

extracted from lemons and oranges.

Table 2: Total Italian exports of citrate of lime and total United States imports

Given the extent of the production and the international demand for lemons, the sector

was of strategic importance for the Sicilian economy.13 As shown in Table A2, the total

(almost 564,559). In total, there were 8,287,758 lemon trees across the peninsula.12These data should to be evaluated with some caution given that the total Italian lemons exports refer to

the calendar year while the total US imports refer to the �scal year (and the two do not coincide).13The leading role of Sicily for the US lemons market continued until the end of the 19th century, when

the production in Florida became substantial.

7

exports revenues from the harbor of Messina in 1850 was approximately 21.6 millions of lire

(at current price). Revenues from citrus and derived products equaled almost 9.2 millions of

lire, accounting for almost 42.4 percent of the total exports revenues. In 1873 the percentage

went up to more than 50 percent of the total exports (Battaglia, 2003).

As mentioned earlier, these substantial revenues are mainly explained by the comparative

advantages Sicily was experiencing because of its geographic location and climatic conditions.

In particular, the island�s hot coastal plains, together with the exceptionally fertile soil made

of limestone base with heavy coatings of lava, were well-suited for growing citrus fruits.

Moreover, lemon trees have a very poor tolerance to extreme climates and to grow and

develop require temperature to be between 13-30 �C, otherwise �owers (and fruits) may die

after few minutes of exposure to temperatures below 1-2 �C. This intolerance to frost explains

the geographic concentration and location of the trees over the island: areas slightly above

the coastline are more suitable because of the relatively low variation in daily (and annual)

temperature, while locations in the mountains with a larger daily (and annual) variation in

temperature are less suitable for cultivation. Also, lemon trees require constant provision

of water. The irrigation system used in the plains surrounding Palermo worked thanks to

the norie, a quite advanced technology that enabled farmers to collect water very deeply

in the soil.14 Yet, given that local institutions often failed to provide an adequate public

irrigation system, local producers had to make costly investments in order to build and

maintain private ones.15

We argue that the high exports revenues from citrus combined with the sizeable initial

investments needed for the development of citrus cultivation, made local producers very

vulnerable to predation. In the absence of public protection, the ma�a exploited this systemic

vulnerability to extort part of the pro�ts made by the industry. Consequently, we consider

pro�ts coming from imperfect market structures as a necessary condition for the development

of the ma�a.

3 The model

The model considers three active agents -households, lemon producers, and the ma�a - as

well as a (latent) government that determines the strength of property rights. The aim of

the model is the demonstrate the mechanism through which a boom in demand caused ma�a

to arise in certain towns but not in others. Speci�cally, we show how the structure of �xed

costs a¤ected the decision of whether to produce lemons or not and, if production occurs,

under what circumstances the lemon producers chose to pay the ma�a for protection against

14The norie were initially moved by animals, but already in the 1872 they started to be moved by steampumps.15Thanks to British merchants, the Sicilian wine industry started in the late eighteenth century around

the area of Marsala (in the province of Trapani). There exist signi�cant di¤erences though between thecultivation of grapes and the cultivation of citrus. First of all grapes are much more tolerant to frost andthey can live even up to 1000m over the sea level. Also, because they are not an ever green plant (like citrus),during winter these plants rest and therefore are in a better position to face extreme weather conditions suchas snow. The area surrounding the province of Trapani was naturally fertile and did not require speci�cinvestments in water irrigation.

8

thieves.16

3.1 Households

Let us assume that the utility of a representative household is given by

U = � lnC + (1� �) lnX

where U is utility, C is consumption of lemons from Sicily, and X is consumption of other

goods.17 The parameter � indicates the taste for lemons within the representative household.

For simplicity, we assume that the island of Sicily has a monopoly in the production of

lemon.18 The representative household might be thought of as the average Western household

in the last decades of the 1800s. The individual household�s budget constraint is Y =

pC + X, where Y is average household income at the time and p is the relative price of

lemon consumption (price of other goods X is normalized to unity).

From the �rst-order conditions for pro�t maximization, we can obtain the inverse demand

function for Sicilian lemons:

p (C) =�Y

C

As usual, there is a negative association between price and the total level of demand C,

whereas demand rises with income and with the preference for lemons �.

3.2 Lemon producers

There are in total I > 0 towns or municipalities in Sicily and 1 < n � I towns where lemons

are produced. For simplicity, we assume that each town has one (representative) producer.

Total supply of Sicilian lemons is C =XI

i=1Ci where Ci is the local level of production in

town i. Total supply always equals total demand.

Pro�t of the local producer in town i is

�i = p (C) � Ci � (Ci)� Fi =�Y

C� Ci � Ci � Fi

where p (C) is the price level (which depends on total demand), (Ci) is a marginal cost

function such that 0 (Ci) = > 0, and Fi is the local �xed cost of entry into lemon

cultivation. Fi depends on local characteristics such as soil quality, water access, altitude,

and slope of the land, as well as non-community speci�c �xed costs such as costs of building

protective walls, etc. Typically, it takes several years before planted lemon trees have grown

to produce lemons. Once a lemon plantation has been established, the marginal cost is

the same across localities.16The model is meant to describe the situation in the 1880s and is not necessarily relevant for understanding

the contemporary nature of ma�a operations.17We will only refer to lemons in the model below, but as indicated in the section above, what we really

have in mind is the market for lemons, oranges, limes, and other citrus products.18This is a simpli�cation. However, Table 2 shows that in the dominant US market, Sicilian lemon accounted

for 100 percent of all imports during certain years in the 1898-1903 period.

9

The �rst-order condition for pro�t maximization can be written as

p (C)

�1 + p0 (C) � Ci

p (C)

�= 0 (Ci) :

Since marginal cost and inverse demand p (C) are the same everywhere, Ci must in

optimum be identical in every town. Hence, C = nCi. The expression above can therefore

be written as�Y

nC�i

�1� 1

n

�=

The fact that the number of towns n � I is bounded from above implies that there will

be a positive mark-up over marginal cost and that the market is not fully competitive.

Solving for the Cournot equilibrium supply of lemon from town i gives us

C�i =�Y (n� 1)

n2 (1)

Not surprisingly, equilibrium supply will increase with the typical income Y and decrease

with marginal cost . Furthermore, it can easily be shown that C�i will decrease with n for

all n > 2.

Inserting C�i back into the pro�t function, we receive after some algebra the optimal

pro�t level

��i =�Y

nC�i� C�i � C�i � Fi =

�Y

n2� Fi:

In this very simple expression, pro�ts increase with income and decrease with the number

of producing towns n. Obviously, lemons will only be produced in community i if ��i =�Yn2� Fi � 0: Hence, �xed costs and the number of other producers are potential barriers to

entry into lemon production.

Let us assume that towns i 2 f1; 2; 3; ; ; Ig are ordered such that F1 < F2 < F3::: < FI .

Let us further assume that �xed costs are uniformly distributed across towns and that they

are simply given by

Fi = a+ bi

where a > 0 is a component common to all towns and where b > 0 is a parameter describing

the gradient of �xed costs across towns. One might for instance think of a as capturing the

cost of building protective walls, which is roughly the same everywhere, whereas b might

capture the di¤erence in �xed costs that arises due to di¤erences in soil quality or access to

surface water for irrigation which make it more costly in terms of time and e¤ort to establish

a lemon plantation in some places rather than in others. Clearly, a b close to zero would

imply small di¤erences between towns. The mean �xed cost across towns is �F = a+ I � b=2.With these assumptions, the last producer who will choose to produce lemon (i = n) will

10

be the one for which the following condition holds:19

��n =�Y

n2� Fn =

�Y

n2� a� bn = 0: (2)

All potential producers i 2 f1; 2; 3; ; ; ng will thus produce whereas i 2 fn+ 1; :::Ig willnot. By using the implicit function theorem, we can deduce from the equation above that

the equilibrium level of lemon-growing towns is a function n = n (a; b) such that

@n

@a= na =

�12�Yn3

+ b< 0;

@n

@b= nb =

�n2�Yn3

+ b< 0:

Although the explicit solution to n is mathematically messy, it is easily illustrated in a

graph as in Figure 1. The �gure shows the two components of the pro�t level, �Y=i2 and

a+ bi; as a function of i when towns are ordered, starting from the one with the lowest �xed

costs to the left. The equilibrium occurs at the point where the two lines cross. At n, pro�ts

for the nth �rm are zero whereas they are given by the distance between �Y=n2 and a + b

for the �rm with the lowest �xed costs.20 The triangle D in the �gure shows the total pro�ts

of the lemon producing sector in Sicily.

Figure 1: Equilibrium number of lemon-producing towns

It is clear from the �gure that an increase in a and/or b would shift the Fi curve to the

left, resulting in a lower n. Over time, it is likely that such barriers to entry have varied

in the lemon trade just as in other sectors. As a thought experiment, one might imagine

another agricultural good (perhaps wheat) with the same pro�t function except that it had

lower barriers to entry al < a and bl < b, as shown in the bottom of Figure 1. Such low

levels of �xed costs would imply that all towns (n = I) would produce the good and that

average pro�ts would be quite small. Total pro�ts in the sector are given by the distance

between the �xed cost curve Fi = al + bli and the pro�t level �Y=I (the area E):

Hence, the individual pro�t for an actual producer is

��i =�Y

n (a; b)2� a� bi � 0 for all i � n:

An increase in the �xed cost coe¢ cients a and b thus has two e¤ects on equilibrium

pro�ts: it reduces the equilibrium number of lemon producers, which has a positive e¤ect

on pro�ts in town i, and leads to an increase in the �xed costs for all producers, which

decreases pro�ts. The sign of the comparative statics will depend crucially on the level of

i.21 In general, for a given n, pro�ts fall with i. Pro�ts always increase with household

demand �Y . We can therefore express ��i = � (�Y; a; b; i).

19 In the expression below, we assume for simplicity that there always exists a level of pro�ts such that��n = 0. In reality, the equilibrium number of lemon-producing towns n� would probably rather be de�nedby n� = argmin max

��Yn2� a� bn; 0

.

20The pro�t level for the 1st �rm is equal to b (n� 1) > 0:21We can for instance see from Figure 1 that a rise in b with a unchanged will increase pro�ts for the town

with the lowest �xed costs, whereas the previous nth �rm will then have negative pro�ts and should cease toproduce.

11

3.3 Government

As described above, Sicily in the 1880s was characterized by weak property rights institutions

and a substantial number of thieves who predated on agricultural production. An implicit

assumption in this section is that the �predation technology� in the lemon business was

particularly favorable for thieves. Compared to other agricultural goods like grapes or wheat,

lemons are very easy to collect quickly by a prospective thief, and the price per stolen bucket

is further very high. These factors contributed to lemon plantations being in particular need

of protection.

Let us assume that in each community there are d > 0 thieves.22 In the absence of

property rights and other forms of protection, thieves would steal the full pro�t from lemon

production and each thief would obtain an amount of ��i =d. Lemon production would then

make zero pro�ts. The government in Rome o¤ers some protection of property rights cap-

tured by the term � 2 [0; 1], where � = 1 implies perfect enforcement of property rights

whereas � = 0 implies total absence of government protection. For Sicily in 1880, � was

presumably closer to 0.

The total proportion of pro�ts saved from thieves by the individual lemon producer is

given by the �predation success function�:

� (mi) =mi

mi + d (1� �)

where mi is the level of private protection o¤ered in i.

The functional form implies that if � = 1, there is no need for private protection since

� (mi) = 1 for any level of mi. If � > 0, then �0 (mi) > 0 and �00 (mi) < 0, i.e., the proportion

of protected pro�ts is a positive, concave function of the level of private protection. Lemon

producers then retain � (mi) � ��i of their pro�ts and lose (1� � (mi)) � ��i to the thieves.Lemon producers cannot provide protection themselves and hence need to employ people

to do this job for them. This is where the ma�a comes in.

3.4 Ma�a

The nature of the original creation of local ma�a groups (cosca) remains a mystery. What we

know about such groups is that they formed a secret society of sworn-in men who probably

grew out of a kind of organized local resistance against Sicily�s foreign regimes. The members

managed to overcome the collective action problem through various measures (like brutal

punishments in the case of defection). Ma�osi were recruited among men with very diverse

occupations in society, including peasants, sheep-herders, doctors, and politicians. In these

early days, ma�osi typically performed their daily jobs as an integrated part of society while

22 In this model, being a thief is di¤erent from being a ma�oso. A thief only steals and on an individualbasis, whereas the ma�a expropriate rewards in an organized manner by o¤ering "protection". In reality, itis not easy to draw a line between the two. In a richer model, the number of thieves might be endogenized sothat individuals self-selected into being a ma�oso, a thief, or a normal peasant in a process where marginalreturns were the same in equilibrium. The number of thieves in Sicily in the 1880s was reportedly very highdue to a general release of prisoners after the Italian uni�cation and the breakup of feudal estates, whichmade many workers redundant.

12

also undertaking ma�a activities on the side. In the latter half of the 19th century, it is

known that the key ma�a activity was the protection of businesses but we do not know if

this was their original purpose (Gambetta, 1996).

In this paper, we take as given that in each town, there is a tight network of individuals

who have formed a secret ma�a association for some unknown purpose. We assume that

ma�a members mainly pursue normal economic activities but that they might switch to

o¤ering organized and coordinated protection to farmers when conditions are exceptionally

favorable, such as during the citrus boom in the mid-1800s. If instead normal economic

activities are more rewarding, the local ma�a will remain a �sleeping cell�in the population

that does not activate an alternative business of organized protection. In our model, the

local ma�a organization in i has no in�uence over n (there was no central coordinating

ma�a authority in the 1880s). A representative ma�oso considers the choice of allocating

e¤ort either to protecting local lemon producers or to pursuing normal economic activity. A

representative ma�oso�s utility function in town i is

UMi = !� (mi)��i + (1�mi)A

wheremi 2 [0; 1] is available e¤ort that can be spent on protecting the local lemon producers�pro�ts. The parameter A > 0 re�ects productivity in normal production (farming, �shing,

herding sheep, etc). This type of production is one option available to ma�osi and is the

only available option for the majority of ordinary people. ! 2 (0; 1) is the share of totalprotected pro�ts that the local producers o¤er to the ma�a in return for protection. For

now, let us take ! as given.23 Note that ! must be somewhere within the interval (0; 1) for

any interaction to occur between the two.

The ma�a maximizes the utility function:

maxmi

UM =!mi�

�i

mi + d (1� �)+ (1�mi)A:

After manipulating the �rst-order conditions, we can solve for the optimal (interior so-

lution) level of ma�a activity in town i:

m�i = d (1� �)

s!��i

d (1� �)A � 1!

(3)

Clearly, m�i > 0 will only be the case if the term under the square root sign exceeds 1.

The expression in (3) implies that we can express the following proposition:

Proposition 1: The ma�a will be active in town i (m�i > 0) only if !�

�i = !

��Y

n(a;b)2� a� bi

�>

d (1� �)A:

This proposition o¤ers some of the key insights of the model. If the opportunity costs of

23Please see Figure A1 in the Appendix for an extension and simulation of the model where !i is en-dogenously determined in a two-stage game between lemon producers and the ma�a. The main comparativestatics results below still hold after this extension.

13

being an active ma�oso A are very large, there will be no protection activity. If the o¤er from

the producers ! is very low, the ma�oso will not �nd protection worthwhile. Furthermore,

it will obviously be the case that there will be no protection if property rights are fully

enforced, i.e. if � = 1. It can be shown that m�i is a decreasing, convex function of � so that

the protection activities shrink as government-enforced property rights are strengthened.

Similarly, there will be no protection if there are no thieves so that d = 0. All these factors

are assumed to be identical throughout Sicily but might explain the varying presence of

ma�a activity over time.

What distinguishes towns is the level of pro�ts in lemon production ��i . The central

result is of course that the likelihood of ma�a activity increases with ��i . As discussed above,

we argue that one of the key distinguishing features of lemon production at the time was the

relatively high demand �Y and the high barriers to entry due to high and geographically

di¤erentiated �xed costs, represented by the parameters a and b. If the latter are high, then

only n < I towns will be able to produce and the average pro�t among these producers will

be relatively high. For other goods, we argue that a and b should be fairly low, implying low

pro�ts in general and no large geographical variation in pro�ts. The lower part of Figure 1

depicts such a scenario. Pro�ts are then less likely to motivate a ma�a to become activated

from (3).

The most likely place for protective activity would be town i = 1 where �xed costs of

lemon production are lowest and pro�ts are highest. In our empirical investigation, we do

not have data on pro�ts from various types of production. What we do have data on is the

presence of sectors in each town. According to our model and the data discussed above,

the presence of lemon production in some towns should be an indicator of pro�tability and

of low �xed costs. Similarly, the presence of other types of production are interpreted as

indicating that pro�ts in that sector were also positive. Holding the presence of other types

of production constant, our hypothesis is thus that the prevalence of lemon production in a

town should have a positive association with the probability of ma�a activity.

4 Econometric speci�cation and data

4.1 Econometric speci�cation

From an econometric point of view, we can consider equation (4) as the latent equation that

will determine the probability of ma�a presence. Ma�a depends on: pro�ts, enforcement

of property rights and the number of thieves. The latter are considered equally distributed

across the region even though the presence of the state (and the consequent strength with

which property rights are enforced) might di¤er at town level, explaining the variation in

ma�a activity across the region. For this reason, the e¢ ciency of the enforcement of property

rights is included among the control variables.

The model to be estimated can be written as:

M?i;p = �p + �1�i;p + �2Zi;p + �i;p (4)

14

where

Mafiai;p =

(1 if M?

i;p � 00 if M?

i;p < 0

In the latent equation (4), the dependent variable M?i;p represents the response variable

that drives the probability of ma�a in town i in province p. A response variable larger than

zero is associated with towns where there is ma�a presence. The probability of ma�a is zero,

if the response variable (M?i;p) is smaller than zero.

The main explanatory variable is the local pro�ts from citrus production, which we denote

by �i;p. Pro�ts depend on �xed costs, which in our model represent a barrier to entry in

the market for lemons. As a proxy for the actual pro�ts, we consider the presence of citrus

production in a town as a signal that �xed costs were low enough to allow for positive pro�ts.

Zi;p represents a set of control variables that may also a¤ect the probability of ma�a such

as the degree of trust citizens have in the law and the peripherality of each town. These

measures are not intended to provide a proxy for the e¢ ciency of the state in enforcing

property rights. Finally, �p represents provincial �xed e¤ects that may be correlated with

the error term �i;p.

4.2 Data

Data both at town and district level, i.e.(mandamento)24, are collected for the seven provinces

in which Sicily was split at the time of analysis (Caltanissetta, Catania, Girgenti, Messina,

Palermo, Syracuse, and Trapani) for a total of 143 observations.25

The section of the Inquiry that matters to our analysis is made of two parts. The

�rst one discusses the situation of the agricultural sector, with particular reference to tax

burden, wages, the kind of crops produced, and the relations between peasants and landlords

(i.e., tenancy contracts, fractionalization of land, etc.). Questionnaires relative to this �rst

part were sent out to almost 357 mayors, but less than half of them provided complete

information.26

The second part of the Inquiry provides information on the moral and social conditions of

peasants. Questionnaires were sent to 179 pretori (lower court judges).27In this section, we

focus on the type/level of crime in the region. The question asked to the pretori is: �What

is the most common form of crime in the district? What are their causes?�We coded as

dependent variable a binary dummy, ma�a, for whether the pretore of the town recognizes

24A mandamento is a judicial district of competence of the pretore.25Syracuse became province in 1865 replacing Noto, which was a province at the time of the uni�cation in

1861. Except for ma�a activity, all the information regarding Caltanissetta comes from the Summary Reportthat Abele Damiani presented to the Parliament because the original handwritten copy of the Inquiry hasnever been catalogued in the Archive of State in Rome.26For unknown reasons, di¤erently from the other provinces, the folder for this section on the province of

Caltanissetta never reached the Archive of State in Rome and therefore, it was never catalogued there. Inorder to get data for the agricultural conditions for this province, we use the information available from theSummary Report that Damiani presented to the Parliament, which has also been used by Bandiera (2003)and Buonanno et al. (2012).27There are much fewer pretori than mayors since the pretura is only present in larger provinces and one

pretore often serves several towns.

15

ma�a as the most important source of crime in the district.28

Bandiera (2003) and Buonanno et al. (2012) opted for a di¤erent dependent variable:

an ordinal variable for the intensity of ma�a collected from the Summary Report that Abele

Damiani wrote to the Italian parliament on the basis of the original Inquiry. However, from

the original Inquiry it emerges that very few pretori mention the intensity of ma�a (10 out

of 143). Since the origin of the additional information is unclear, we prefer to stick to the

original document, for which the source appears to be more accurate.29

There are potential concerns with the data on ma�a presence. First, could the ma�a still

be present in a district even though the pretore did not list it as the most common form

of crime? Because of the structure of the survey in the Inquiry, it is indeed possible that

some districts had ma�a activity, even though the pretore did not report it as being a major

crime. This problem may slightly a¤ect our results.

Second, were pretori themselves ma�osi and hence, likely to understate the presence of

ma�a? The answer to this question is most likely to be no (although there is no conclusive

evidence in either direction). Pretori were directly appointed by the Minister of Justice with

a Regio decreto (royal order) and any other aspect concerning their career was subject to

an evaluation made by a committee of experts belonging to the local Court of Appeal. For

the �rst 10 years of their career, pretori changed district (mandamento) very often, which

may have restricted their possibilities of connecting with the local environment.30

Third, did the pretori have a general understanding of what the term ma�a implies?

This indeed appears to have been the case. In the 1880s, the word ma�a was already used

to indicate a criminal organization, at least since 1863, when a comedy titled �I Ma�usi di

la Vicaria�was shown in Palermo. Also, in 1865 the prefect of Palermo (Filippo Gualterio)

used the word ma�a in a private document to identify the criminal organization and lately,

in 1871 ma�a membership became a public law o¤ence. We therefore regard it as highly

unlikely that there was a misinterpretation. 31

Figure 2 shows the local distribution of ma�a in our sample. On average, the 36 percent

of towns were strongly a¤ected by ma�a, which means that almost 51 out of the 143 towns

had ma�a listed as the most common form of crime. Girgenti is the province with the highest

presence of ma�a, where 14 out of 17 towns have strong ma�a presence. In Trapani, the

ma�a operates in 6 out of 15 towns, and in the Caltanissetta province in 7 districts out of 16.

28Data on ma�a are available for 162 districts, but when merged with the independent variables the largestsample covers 143 districts. See Figures A2 and A3 for examples of excerpts from the original inquiry.29Pretori were lower court magistrates. Because of their role, their information on criminal activity can be

considered the most indicative.30From another question of the survey it emerges that many pretori complained about administering justice

without the cooperation of locals, since in several trials witnesses did not testify because of ma�a�s retaliationor collusion with the ma�osi. According to Pezzino (1990), the pretore in Bagheria said �There is a tendencyto deny the truth. Not only people does not answer truthfully, but they deliberately lie either because ofma�a or because of money or because they are scared.�31As long as the error term is not correlated with independent variables, there is no reason to believe that

these problems will a¤ect estimates. This is important for our analysis given that the possible misinterpre-tation of the question as well as the collusion between pretori and ma�a (if any) are likely to be distributedrandomly across towns with ma�a. It seems reasonable then to assume that the independent variable that webase our analysis on (whether a city/village produces citrus) is not correlated with the measurement error,providing us with unbiased estimates.

16

Almost one third of the districts in Palermo (mainly those in the Conca d�Oro) and Catania

provinces are ma�a infested. Messina and Syracuse are the ones with the lowest incidence.

These statistics are also consistent with Colajanni (1885).32

Table 3: Distribution of ma�a and agricultural production across provinces in1881-86

Figure 2: Ma�a and non-ma�a towns in Sicily in the 1880s

For this reason, the set of independent variables that we employ in the analysis can be

divided into three groups. Colajanni (1885), Dickie (2004), and Lupo (2009) identify the

pro�table production of goods as important determinants of ma�a presence. For this reason,

the �rst group of independent variables (see Table 3) includes Citrus, Wheat, Olives, Grape

and Sulphur. 33 Given that the Inquiry does not provide information on hectares per crop

for every mandamento, we decided to use a dummy variable (only recording whether the crop

is predominant or not) in order to minimize the potential measurement error. The question

we draw data from in the Damiani Inquiry is: �Which is the dominant crop produced in the

city?�. Mayors listed more than one crop (for few cities they also reported quantities), and

because of that dummies are not mutually exclusive.

Buonanno et al. (2012), on the other hand, opt for a measure of crop suitability to

proxy citrus, arguing that it should be preferred to actual data on crop production because

of possible reverse causality. Even though this is generally true, there are at least two

reasons why in this case a measure of actual production is preferable (given that there are

data available for it). First, the risk of reverse causality between ma�a presence and citrus

production in this occasion is minimal: production of citrus in Sicily responded to an external

shock in the demand related to the use of the fruit as a treatment for scurvy. As argued

by Baron (2009), this shock happened in the late 1700s (after Lind�s discovery) while the

origins of ma�a have been usually dated around the 19th century.

The second reason relates to the fact that crops suitability can be signi�cantly altered

by investments in irrigation. As argued by Pizzuto Antinoro (2002), locations that appeared

to be unsuitable for production because of soil characteristics were transformed into being

productive by investments in water irrigation and similar technologies (for example, the norie

previously mentioned). But these investments, driven by the possibilities of high revenues,

made also producers vulnerable to violence, predation and to ma�a retaliation.

Because data on citrus production play a key role in our analysis, in a couple of instances

where the Inquiry has ambiguous information, we used the data available in Di Vita (1906)

32Using information available in Damiani (1886), Colajanni identi�es three macro-regions on the basis ofeconomic and social conditions. The �rst region includes the province of Catania and Messina, where therewere good economic and social conditions. The situation was slightly worse in the provinces of Syracuse,Trapani, Caltanissetta, and Palermo. The third region includes Girgenti, where both economic and socialconditions were very poor and there was a high level of crime.33The full list reported in the Inquiry for all the Sicilian region has many more crops and plants but we

decided to include only 3 other than citrus on the basis on their relevance.

17

to complement the Damiani Inquiry. 34Di Vita (1906) also provided data on sulphur mines

and we used them to code this new variable.35As argued by Colajanni (1885), sulphur mines

are almost exclusively concentrated in the province of Girgenti (12 out of 17 towns). Outside

Girgenti, there are 5 mines in the province of Catania, 3 in Palermo, and 1 in the province of

Messina and Trapani. Wheat production instead is high in the entire province of Girgenti,

but is low in the province of Messina. Grapes and olives are almost equally distributed across

the island. Our summary statistics seem to match the picture provided in Colajanni (1885)

quite well.

The second group of explanatory variables intends to control for the political status of

each town and to assess the impact of policies implemented between 1812 and 1870 for

increasing the small-scale ownership of land (see Table A3 in the Appendix). We consider

three types of policies: i) the abolition of feudalism and the auction/allocation of land to

smallholders; ii) the en�teusi, a perpetual lease that allowed farmers to use the land as if they

were owners; and iii) the seizure of church-ruled territories and the consequent land auctions,

which occurred after Italy�s uni�cation. For each city, mayors provide information on the

e¤ectiveness of these policies in increasing land fragmentation. We use this information to

code a dummy variable which is positive in case of reported e¤ectiveness. According to our

data, highest e¤ectiveness was reached in the Caltanissetta, Girgenti, and Catania areas,

where the latifund was dominant. Yet, in the provinces of Palermo and Caltanissetta, the

distribution of land appears to have been more fractionalized in relative terms. This was

mainly due to the fact that the majority of cities in these provinces were directly ruled

by the crown, preventing the typical feudal hierarchy, dominant in other provinces. As a

consequence, fractionalization policies had little e¤ect in increasing private ownership among

peasants.

Regarding land distribution, we note that fractionalization is high in almost 50 towns

and relatively low in 45 towns.36The questions asked are: �What is the dominant scale of

the plantation? And what is the fractionalization of land?�Table A4 shows the distribution

in detail. The scale of plantation tends to be relatively high in Palermo, whereas in the other

provinces its percentage is around 33 percent. Small-scale plantations are fairly common in

all provinces, but particularly in Trapani, Messina, and Catania. Girgenti and Caltanissetta

instead are the provinces with the lowest number of small-scale plantations.37

All our independent variables are weakly only cross-correlated, preventing problems of

multicollinearity (see Table A5 in the Appendix). The variables are described more closely

in Table 11.34 In fact, in few istances, the Inquiry reports that the lemons� industry is relevant in a speci�c city but

then does not list lemons among the relevant crops. These cases were compared with Di Vita (1906) andcoded according to this latter source.35See Figure A4 in the Appendix for a more recent distribution of crops and fruit trees in Sicily.36The Damiani Inquiry (1886) is the source for these data, except for Caltanissetta, whose data come from

Damiani�s Summary Report.37Most of the time, mayors answered that a large, a medium, and a small scale are dominant, therefore the

sum of the three variables is larger than 1.

18

5 Empirical analysis

5.1 Ma�a presence in the 1880s

Table 4 presents probit estimates with ma�a presence in the 1880s as the dependent variable.

We start by estimating a simple model, where ma�a presence in the 1880s depends only on

variables capturing the economic activity of the town. We then proceed by introducing

additional variables to control for observables. In Column 1, the di¤usion of the ma�a

signi�cantly depends on citrus production. At the mean, the production of citrus increases

the probability of ma�a by 45 percent. In Column 2, we re-estimate the same model but

controlling now for province dummies to capture regional �xed e¤ects: results still hold. In

Column 3, we drop the observations for the province of Caltanissetta in order to detect any

potential bias due to a di¤erent source of information. The estimated e¤ects in Column 3

are almost unchanged. Coe¢ cients and z-statistics su¤er from negligible variation.

In Column 4, we change speci�cation by dropping the non-signi�cant variables (to prevent

an excessive reduction of the degrees of freedom) and by entering two additional controls.

We use a Fractionalization Policy dummy and the variable Population density to account for

policies that may have a¤ected the private ownership of land and wealth. Population density

is not signi�cant, whereas the Fractionalization Policy dummy has a strong signi�cant e¤ect

on the probability of ma�a presence.

Bandiera (2003) argues that the e¤ect of fractionalization policies occurs through land

fractionalization. However, it is also possible that the dummy for fractionalization policies

captures the absence of public providers of protection (i.e. the lord) after the abolition of

feudalism. In fact the aim of these policies was to limit the power of lords by redistributing

land to private owners. However, even though private ownership did increase, the resulting

fractionalization in former feudal cities never exceeded the one in crown-ruled cities where

land has always been fractionalized due to the absence of lords. Therefore, to better assess

the e¤ects of this last control, in Column 5 we also include a dummy variable for the degree

of land fractionalization. The dummy variable High land fractionalization turns out to be

non-signi�cant. As argued above, though fractionalization policies in former feudal cities

had some e¤ects in increasing private ownership, the overall e¤ect was not large enough and

therefore, land distribution was not more fractionalized than in former crown-ruled cities.

The large �xed costs related to investment in irrigation (norie) were much more likely

to be sustained in towns where the scale of the plantation was relatively large (because of

the decreasing cost per hectare), making producers more vulnerable to a potential loss due

to extortion. To test this hypothesis, Column 6 includes the dummy variable Large-scale

plantation. The variable turns out to be highly signi�cant, increasing the probability of

ma�a presence by 39 percent. The set of covariates speci�ed in Column 6 represents our

preferred speci�cation. In this model, the presence of ma�a is signi�cantly determined by

the citrus production, the e¤ect of policies for private ownership, and by the scale of the

plantation.

Table 4: Ma�a Probit Model

19

In Table 5, we proceed by testing the robustness of our preferred speci�cation. We start

by including, in Column 1, control variables to account for the distance from Palermo and

from Mazzara del Vallo. The �rst variable should capture the distance from a well-known

ma�a-base center and also a key port, and the variable Distance from Mazzara del Vallo

is used to capture the possible di¤usion of citrus. In the 10th century, lemon plants were

originally introduced by the Arabs from Mazzara del Vallo. The former variable turns out

to be non-signi�cant, whereas the latter is marginally signi�cant, i.e., the presence of ma�a

decreases if the distance from Mazzara del Vallo increases.

In Column 2 we introduce the dummy variable Feudal to test whether this political sys-

tem, often associated with patronage and kinship relations, had any impact on the presence

of ma�a. The variable turns out to be non-signi�cant.

In Columns 3 and 4, we control for the reach of local governments by using the distance

from the nearest railway station (collected from Di Vita, 1906) as a measure of peripherality

and three dummies (from the Damiani Inquiry, 1886) to account for whether citizens trust,

mistrust, or do not care about the law (the excluded group is whether they fear the law). The

results show that being close to the railway station has a negative e¤ect on ma�a presence

(consistent with the peripherality hypothesis). Towns where citizens do not care about the

law have almost 60 percent higher probability of ma�a presence.

Finally, in Column 5 we control for the length of the tenancy contract which also turns

out to have a signi�cant and negative e¤ect on the probability of ma�a presence. The Citrus

dummy still increases the probability by 57 percent and the Fractionalization policies dummy

increases the probability by an average of 23 percent.

Table 5: Ma�a Probit Model with additional controls

5.2 IV and causality

As discussed above, our hypothesis is that the positive shock on the demand for citrus,

following Lind�s discovery of the bene�cial properties of citrus fruits in the treatment of

scurvy, together with a comparative advantage in climatic conditions, gave Sicily a dominant

position in the market for lemons. This in turn resulted in larger pro�ts for some Sicilian

producers in a weakly institutionalized setting, which created a demand for the ma�a. The

historical evidence from exports of citrus from Sicily and prices of lemons in the 19th century

provided in Section 3 both o¤er support to this hypothesis.

However, measurement errors and omitted variables represent, as always, a possible threat

to the identi�cation of the e¤ect. For this reason, in this section we re-estimate our preferred

speci�cation using an IV estimator in order to determine whether the documented correlation

between lemons and ma�a is in fact causal. The IV provides a source of exogeneity to our

dummy for citrus fruits, which allows us to deal with both the biases above. In addition, an

IV estimator, providing more variance, will also allow us to overcome the problem that the

independent variable is a dummy, given that the estimator exploits the covariance between

dependent and excluded instrument.

20

To provide some exogeneity to the dummy for citrus fruits, we use data on thermal

regimes from FAO GAEZ38. Among the several indicators on thermal regimes provided by

the FAOGAEZ, we chose a measure of the frost-free period.39 GAEZ estimates of climate and

agro-climatic analysis are based on mean climatic data for the period 1961-1990. Therefore,

we assume that drastic changes in climatic conditions have not occurred during the last

two centuries. The reason for using data on the risk of frost as an instrument for lemons

is related to the minimal tolerance of the lemon trees to frost: the probability of frost,

therefore represents an important �xed cost for the production of lemons. We claim that

lemon production will occur in towns characterized by a mild climate with a shorter frost

period and a lower seasonal variation. High pro�ts will be generated by this activity and

more protection against potential losses will be required from local producers, playing a key

role on the level of ma�a activity.

The equations we estimate can be written as follows:

Mafiai = p + �1Lemoni + �2Xi + �i (5)

Lemoni = �p + 1Frosti + �2Xi + �i (6)

The �rst stage regression (eq. 6) predicts the probability of producing lemons, which

then, in the second stage, a¤ects the probability of ma�a (eq. 5).

In Table 6, Panel A, we show results for the second stage regression. The coe¢ cients

and levels of signi�cance for the excluded instrument, together with diagnostic tests are also

reported. In Column 1, we report estimates for the IV and all variables in the baseline model

are statistically signi�cant at least at the 5 percent level. Citrus increases the probability

of ma�a presence by almost 57 percent, Fractionalization policies by almost 24 percent, and

Large-scale plantations by almost 19 percent. Diagnostic tests con�rm the relevance of the

instrument. The Cragg Donald F-statistics is well above critical values for weak instruments

and the partial F-statistics from the �rst stage is well above 10 which is normally considered

the threshold for relevance of instruments with one endogenous variable (Stock and Yogo,

2005). We also report estimates from using an IV Probit (Column 2) even though this

estimator is considered to have identi�cation problems providing biased results. However,

estimates in Column 2 are consistent with IV estimates.

Table 6: 2-SLS Estimates

To provide more support to the validity of our instrument, we decided to undertake a

number of falsi�cations tests. We start from the assumption that the main e¤ect of frost is on

agricultural production in general. Therefore, the instrument (Frost-free Period) should have

an e¤ect on ma�a only through citrus (which means all other channels related to production

of other agricultural crops should be excluded).

38Global Agro-Ecological Zones - GAEZ: http://www.fao.org/nr/gaez/en/39The Frost-free period dataset refers to the number of days during the year with low risk of early and late

frosts (days with Tmean > 10�C.). This dataset is the result of the calculation procedures of GAEZ ModuleI (Climate data analysis and compilation of general agro-climatic indicators).

21

Speci�cally, in Table 7 we use a probit model to regress our proxy for free-frost period

against production of grape, olives, wheat, and almonds which were the dominant crops in

Sicily in the 19th century. We �nd an insigni�cant e¤ect of the excluded instrument on the

probability of producing any of these crops (see Table 7). In our view, this �nding strongly

supports the validity of our identi�cation strategy. Among all the crops produced in Sicily

in the 19th century, we only �nd a signi�cant relationship between citrus and frost period,

which seems to exclude other potential agricultural channels.

Table 7: Falsi�cation Tests

6 Robustness check

To test further the robustness of our empirical speci�cation, we now run the same regressions

but using di¤erent data sources both for the dependent and the independent variables.

6.1 Dependent variables

To proxy ma�a we now use data on ma�a intensity provided by Cutrera (1900). Using

information from police o¢ ces and newspapers, Cutrera drew a detailed map of the intensity

of ma�a in Sicily in 1900 (see Figure A4). Therefore, for each town, we code the intensity of

ma�a activity using an ordinal variable ranging from zero (no ma�a) to three (high intensity

of ma�a).

However, there are a few problems with this source. First, this data source records the

level of ma�a in 1900, i.e. more than 20 years later than the Damiani�s Inquiry. During

these two decades, it is reasonable to assume that the ma�a spread for reasons di¤erent from

the ones that determined its emergence (for example, for internal con�icts).40Second, as it

emerges from the map in Figure A4, Cutrera�s data cover both towns with and without ma�a

only for East Sicily (provinces of Messina and Siracusa). For the remaining provinces (i.e.

Palermo, Agrigento, Trapani), the author reports only towns with ma�a, without providing

any information about the ones that are missing (i.e. whether they are not mentioned because

there is no ma�a or because there is no information available).

Because we cannot disentangle between these two motivations, we decided to use two

di¤erent coding strategies: i) We only code a variable for ma�a intensity for towns reported

in the map following Cutrera�s coding rule; ii) we use the spatial distribution of ma�a in order

to interpolate levels of ma�a activity for towns where information have not been reported.

We assume that towns that are in the neighborhood of others with a high ma�a intensity are

likely to be also a¤ected by some sort of ma�a activity. As a result, we can use an inverse

distance matrix to interpolate data for those towns for which there is no information on

ma�a activity. The level of ma�a activity will depend therefore both on the intensity and on

the distance from neighbor cities where ma�a was present. The results of this interpolation

are shown in Figure 3. White areas denote towns with a low intensity of ma�a (between 0

40See Gambetta (1992) for a discussion of the di¤usion of ma�a in the period 1880-1900.

22

and 0.69), whereas black areas denote towns with a high intensity of ma�a (between 2.25

and 3).

Figure 3: Ma�a interpolated using an IDW

6.2 Independent variables

Notwithstanding the potential issues associated with FAO GAEZ data on crop suitability

we earlier refer to, we decided to use them for robustness tests. We collect data for the

three main crops produced in Sicily (lemon, olive and wheat).41 In Figure 4, we overlay

maps for the distribution of ma�a and suitability for citrus (Agro-Climatic Suitability and

Productivity in Tons per hectare). Black areas have a relatively high suitability for citrus

(suitability = 1), whereas white areas have a low suitability for citrus (suitability = 0).42As

Figure 4 shows, the majority of towns with no ma�a (small black rectangles) are distributed,

on average, in regions where there is a relatively low suitability (white areas) for citrus. On

the other hand, regions with a high suitability for citrus (i.e. Agrigento, Trapani and part

of Palermo) have a quite high frequency of ma�a. However, there is a quite large white area

(low suitability) spanning from Messina to Palermo with a high frequency of ma�a and low

suitability.

Figure 4: Ma�a intensity and ma�a suitability

To minimize the risk of omitted variable bias, we use a large set of climatic and geo-

graphical indices related to factors which may a¤ect agricultural production, hence ma�a.

From the FAO GAEZ, we include indices for the median altitude, inland water scarcity,

natural soil fertility (natural soil nutrients), soil workability constraints, and a climatic indi-

cator of net primary production (NPP).43 We complete the list of new independent variables

by adding: i) spatial data on distance from the coast (source the NASA Ocean Biology

Processing Group44); ii) data on soil neutrality (pH) from the (FAO Geonetwork 45)46.

6.3 Results

Table A6 shows descriptive statistics both for the dependent and independent variables. The

mean intensity of ma�a using Cutrera�s coding rule is 1.4 (over a 0-3 scale) and the number

of observations is 289. The mean intensity drops to 1.3 (over a 0-3 scale) when interpolating

ma�a for missing towns with the inverse distance weight matrix. With respect to crop

suitability, the mean suitability for citrus is 0.24 (over a 0-1 scale), which turns to be much

41Data on grape are not provided and because of that we cannot control for grape.42The index has been rescaled on a 0-1 scale where 0 means minimal suitability and 1 maximal suitability.43NPP is normally used as a measure of vegetation and it depends on solar radiation and soil moisture at

the rhizosphere44http://oceancolor.gsfc.nasa.gov/DOCS/DistFromCoast/45Geo-Spatial data from the FAO GeoNetwork are available from:

http://www.fao.org/geonetwork/srv/en/main.home46As expected soil PH is quite largely correlated with soil fertility (correlation = 0.5) given that it captures

one of the main factor of soil fecundity.

23

smaller than the suitability for olives (0.65) and for wheat (0.56). This seems consistent with

our hypothesis that the variation in the production of lemons is likely to be larger because

of the plant�s speci�c needs.

Table 8 shows the results. The dependent variable used in the �rst three columns is the

mean intensity of ma�a using Cutrera�s coding rule. Because of that, the sample is con�ned

to the 289 towns for which the author provides data. The �rst model is estimated using an

ordinal probit and the variable proxying the suitability for citrus is signi�cant at 1 percent

level. In Column 2, we re-estimate the same model using an OLS estimator and in Column

3, we control for spatial correlation of the error terms using Conley (2008) spatial HAC

estimator. For both models, the variable for suitability for citrus turns out to be signi�cant

at 5 percent level at least and its standard error decreases quite signi�cantly, when we control

for spatial correlation.

In the last two models, we change the dependent variable and use the interpolated mea-

sure of ma�a intensity, which allows us to expand the sample to the entire population of

Sicilian towns. We regress this new measure of ma�a intensity against the same independent

variables using an OLS (Model 4) and a spatial HAC estimator (Model 5). In both models,

citrus suitability has a signi�cant and positive e¤ect on the intensity of ma�a. In addition,

the coe¢ cient on citrus suitability remains quite stable (close to 1.2) compared to the our

previous smaller sample. Overall a one standard deviation in suitability to citrus increases

the intensity of ma�a by almost 0.17.

Table 8: Ma�a Intensity and Suitability to Citrus

7 Ma�a persistence

In the 140 years since its origin, ma�a activity has spread all over Sicily. Repeated con�icts

among clans have further a¤ected its territorial distribution. However, the districts where

ma�a originally developed still remain hot spots for ma�a activity nowadays.

Table 9 shows data, at province level, on ma�a murders reported to the police in 1999

according to the 416 bis article of the Italian penal law and also data on the number of asset

con�scated from ma�osi by the Italian government during 1997-2008. 47 Exception made for

the province of Palermo, where numbers are in�ated being Palermo itself the most relevant

hub for the Sicilian ma�a�s activity, it is worth noticing that Trapani and Agrigento (former

Girgenti), that used to be the two provinces with the largest presence of ma�a according to

the Damiani Inquiry (1886), are still today the two provinces with the highest number of

per capita assets con�scated.

Table 9: Ma�a persistence, 1880-2008

In order to get a better idea of the persistence, we estimate the impact of our dummy

variable for ma�a presence in the 1880s from Damiani (1886) on the per capita number of

47The 416 bis article prescribes 3-6 years of prison for members associated with the ma�a criminal organi-zation.

24

assets con�scated in 1999 from the ma�a at town level. As Table 10 shows, the presence of

ma�a in the 1880s explains a signi�cant portion of the variation in the dependent variable.

Towns with ma�a in the 1880s have a signi�cantly higher per capita number of assets con-

�scated in the 1997-2008 period (the marginal e¤ect is almost 0.38 as seen in Column 1).

After controlling for a full set of province dummies (Column 2), the result still holds. In our

view, this suggests a strong persistence of ma�a presence over one and a half century.

Table 10: Historical ma�a presence as a determinant of current ma�a activity

8 Conclusions

Understanding how socially ine¢ cient institutions arise and persist is currently a very in-

triguing research topic in economics. In this paper, we have developed a market structure

hypothesis to investigate the origins of the Sicilian ma�a. Unlike existing works that empha-

size political and historical factors, our analysis identi�es the importance of an exogenous

shock in the international demand for citrus fruits, which led to very high pro�ts in certain

localities, together with the presence of high �xed costs associated with lemons production,

as sources of market imperfections. The extraordinary revenues that certain producers ex-

perienced, as well as the weak property rights and rule of law, provided an ideal breeding

ground for the ma�a.

In the empirical analysis, using data from a parliamentary inquiry of the 1880s, we show

that the presence of ma�a is strongly related to the production of citrus fruits. The results

continue to hold when we include several control variables, address a possible endogeneity

issue and employ two di¤erent dependent variables collected and coded from a later source.

We also �nd that the historical presence of ma�a is a strong predictor of current ma�a

activity.

We believe that our results contribute to shedding light on the emergence of one of the

most harmful economic institutions in Europe during the modern era. Additional studies

on country or regional level along the same lines could potentially uncover the micro level

mechanisms for how and why other ine¢ cient institutions arise and often persist for extended

periods.

25

References

[1] Acemoglu, D. and J. Robinson, 2012. Why Nations Fail: The Origins of Power, Prosperity, and

Poverty. New York: Crown Publishing Group.

[2] Acemoglu, D., S. Johnson, and J. Robinson, 2006. �Institutions as a Fundamental Cause of

Long-run Growth�chap 6 in Handbook of Economic Growth, Vol 1, Part A, 385-472.

[3] Acemoglu, D., T. Verdier, and J. Robinson, 2004. �Kleptocracy and Divide-and Rule: A Model

of Personal Rule�, Journal of the European Economic Association 22(2-3), 162-192.

[4] Agenzia del Demanio, 2008. Beni Con�scati Alle Ma�e, Minister of Home A¤airs Report.

[5] Alongi, G. 1886. La ma¢ a nei suoi fattori e nelle sue manifestazioni: studio sulle classi pericolose

della Sicilia, Bocca Editori, Torino.

[6] Anderson A., 1995. Organized Crime, Ma�as and Governments, in Fiorentini, Gianluca and

Peltzman, Sam (eds), The Economics of Organized Crime, Cambridge, Cambridge University

Press.

[7] Arlacchi Pino, 1986. Ma�a Business: The Ma�a Ethic and the Spirit of Capitalism, London:

Verso.

[8] Bandiera, Oriana, 2003. Land Reform, the Market for Protection, and the Origins of the Sicilian

Ma�a: Theory and Evidence. Journal of Law, Economics, and Organization , 19(1), 218-44.

[9] Baron, Jeremy H. (2009). Sailors�Scurvy Before and After James Lind - A reassessment. Nutri-

tion Reviews: 67(6):315-332.

[10] Battaglia, Rosario, 2003. L�ultimo Splendore: Messina tra Rilancio e Decadenza, Rubbettino

Editore.

[11] Bonfadini, Romualdo, 1876. Relazione della Commissione Parlamentare, Inchiesta sulle Con-

dizioni Sociali ed Economiche della Sicilia, Roma.

[12] Buonanno, Paolo, Ruben Durante, Giovanni Prarolo, Paolo Vanin, 2012. �On the Historical and

Geographical Origins of the Sicilian Ma�a�, Munich Personal RePEc Archive.

[13] Catanzaro, Raimondo, 1992. Men of Respect: A Social History of the Sicilian Ma�a, New York:

The Free Press.

[14] Colajanni, Napoleone, 1885. La Delinquenza della Sicilia e le sue Cause, Giornale di Sicilia.

[15] Colajanni, Napoleone, 1895. Gli Avvenimenti di Sicilia e le Loro Cause, Remo Sandrom, Palermo.

[16] Colajanni, Napoleone, 1900. Nel regno della ma�a, Remo Sandrom, Palermo.

[17] Congdon Fors, H. and O. Olsson, 2007. Endogenous Institutional Change After Independence,

European Economic Review 51(8), 1896-1921.

[18] Cutrera, Antonio, 1900. La Ma�a e i Ma�osi, Paermo, A Reber.

[19] Damiani, Abele 1886. Atti della Giunta per l�Inchiesta Agraria e sulle Condizioni della classe

Agricola, Forzani Tip. Del Senato tomo 2 fasc 4 (Archivio Centrale Roma).

26

[20] Di Vita, Giuseppe, 1906. Dizionario Geogra�co dei Comuni della Sicilia e delle Frazioni Comu-

nali: Con Notizie Storiche, F. Pravatà.

[21] Dickie, John, 2004. Cosa Nostra: A History of the Sicilian Ma�a, Palgrave Macmillan.

[22] FAO/IIASA, 2010. Global Agro-ecological Zones (GAEZ v3.0). FAO, Rome, Italy and IIASA,

Laxenburg, Austria

[23] FAO/IIASA Geonetwork- Geo-Spatial Data; http://www.fao.org/geonetwork/srv/en/main.home

[24] Finley M.I., Mack Smith, Denis, Duggan, Christopher; 1987. A History of Sicily. Elisabeth Sifton

Books, Viking Penguin INC.

[25] Fiorentini, Gianluca; 1999. Organized Crime and Illegal Marteks. Dipartimento di Scienze

Politiche. Universita�di Bologna�Wp 8400.

[26] Franchetti, L., 1876. Condizioni politiche ed amministrative della Sicilia Vol.I, in La Sicilia nel

1876, Barbï¿12 ra, 1877.

[27] Fukuyama, F., 2000. Social Capital and Civil Society, IMF working paper WP/00/74.

[28] Gambetta, D. and Reuters, P. 1995. Conspiracy among the many the ma�a in legitimate in-

dustries in Fiorentini and and Peltzman, Sam (eds), The Economics of Organized Crime,

Cambridge, Cambridge University Press, pp.116-136.

[29] Gambetta, Diego, 1996. The Sicilian Ma�a: the Business of Private Protection, Harvard Uni-

versity Press.

[30] Gambetta Diego, 2009. Codes of the Underworld: How Criminals Communicate. Princeton

University Press.

[31] Grossman, Herschel I., 1995. Rival Kleptocracies: The Ma�a versus the State, in Fiorentini,

Gianluca and Peltzman, Sam (eds), The Economics of Organized Crime, Cambridge, Cam-

bridge University Press.

[32] Hardin, Russell, 1999. Social Capital. In Alt, James E., Margaret Levi, Elinor Ostrom: Compe-

tition and Cooperation. New York: Russell Sage Foundation, 170-189.

[33] Hess, Henner, 1973. Ma�a and Ma�osi: The Structure of Power, Farnborough: Saxon House.

[34] Hess, Henner, 1998. Ma�a and Ma�osi: Origin, Power and Myth, NYU Press.

[35] Italian National Institute of Statistics - ISTAT, 1999. Delitti Denunciati dalle Forze di Polizia

all�Autorita�Giudiziaria.

[36] Levi, M., 1996. Social and Unsocial Capital: a Review Essay of Robert Putnam�s Making

Democracy Work. Politics and Society, 24: 45-55.

[37] Lupo, Salvatore, 2009. History of Ma�a, Columbia University Press.

[38] Mehlum, Halvor, K-O Moene, and R. Torvik, 2006. �Institutions and the Resource Curse�.

Economic Journal, 116(508), 1-20.

[39] Monroe, Seymour Will, 1909. The Spell of Sicily: The Garden of the Meditarranean, Boston,

The Page Company.

27

[40] Montanelli, Indro, 1949. Interview with Don Calo Vizzini. Corriere della Sera, October 30.

[41] Newton, Kenneth, 2001. Trust, Social Capital, Civil Society, and Democracy, International

Political Science Review April 2001 vol. 22 no. 2 201-2.

[42] Pescosolido Guido, 2010. L�Economia Siciliana nell�Uni�cazione Italiana, paper presented in the

Convegno Nazionale di Studi: La Partecipazione della Sicilia al Movimento dell�Uni�cazione

Nazionale.

[43] Pezzino Paolo, 1990. Una certa reciprocita�di favori. Ma�a e modernizzazione violenta nella

Sicilia postunitaria. FrancoAngeli.

[44] Pizzuto Antinoro, M, 2002. Gli Arabi in Sicilia ed il modello irriguo della Conca d�Oro, Palermo:

Regione siciliana, Assessorato agricoltura e foreste.

[45] Powell, Harold G., 1909. Italian Lemons and Their By-Products, Bulletin No. 160, Bureau of

Plant Industry, US Department of Agriculture.

[46] Putnam R., 1993. Making Democracy Work: Civic Tradition in Modern Italy. Princeton Uni-

versity Press. Princeton.

[47] Reuter, P. (1987). Racketeering in Legitimate Industries: A Study in the Economics of Intimi-

dation, Santa Monica, CA: The Rand Corporation.

[48] Sala-i-Martin, X. and A. Subramanian (2003) �Addressing the Natural Resource Curse: An

Illustration from Nigeria�NBER Working Papers, No 9804, NBER.

[49] Sonnino, Sidney and Leopoldo Franchetti, 1877. La Sicilia nel 1876, Firenze, Tip. di G. Barbera.

[50] Skaperdas, S., 2001. The Political Economy of Organized Crime: Providing Protection When

the State Does Not, Economic of Governance 2: 173-202.

[51] Skaperdas, S. and Syropoulos, C. 1995, �Gangs as Primitive States�, in Fiorentini, Gianluca and

Peltzman, Sam (eds), The Economics of Organized Crime, Cambridge, Cambridge University

Press.

[52] Travelyan, Raleigh, 2001. Principi Sotto il Vulcano. Storia e Leggenda di una Dinastia id Gat-

topardi Anglosiciliani dai Borboni a Mussolini, Milano BUR.

[53] van der Ploeg, Frederick, 2011. �Natural Resources: Curse of Blessing?�. Journal of Economic

Literature 49(2), 366-420.

[54] Villari, Pasquale, 1875. Lettere Meridionali, L�Opinione.

28

Figure 1: Equilibrium number of lemon producing towns

αY/I E

(al+bl)

a+b

D

(Fi=al+bli)

αY/n2=a+bn

αY

αY/i2 Fi=a+bi

I 1 n

Figure 2: Mafia and non-mafia towns in Sicily in 1880s (Damiani sample)

Figure 3: Mafia interpolated using an IDW

Note: Small black circles denotes areas with no mafia. Small yellow circles denotes area with mafia intensity equal to 1, equal to 2 for red medium circles, and equal to 3 for

green large circles. The intensity of mafia after the interpolation is denoted on a same scale from 0 to 3 with white areas denoting towns with no mafia and black areas

denoting regions with high intensity of mafia. Source: Cutrera (1900)

Figure 4: Mafia intensity and citrus suitability (productivity in tons per hectare)

Note: White areas denotes towns with a low suitability (suitability = 0) while black areas show regions with high suitability (suitability = 1). The index has been rescaled on a

0-1 rang

Table 1: Total Italian exports of lemon and total United States imports

Total Italian exports

Total US imports

Year Quantity (pounds)

Value (US $)

Exports to the US (%)*

Quantity (pounds)

Value (US $) Imports from Italy

(%)**

1898 325,504,061 3,419,486 41.3 133,374,95 2,521,985 100^

1899 359,473,041 3,234,489 36.7 298,634,448 4,399,160 44.1

1900 331,563,577 3,000,286 29 159,384,389 3,655,926 60.3

1901 368,801,294 3,328,610 29.2 148,334,112 3,516,877 72.5

1902 490,033,260 3,432,677 35.3 162,962,091 3,318,909 100^

1903 459,622,020 3,218,948 31.2 152,775,867 3,087,244 93.8

*Percentages provided by Powell (1909) ** Percentages estimated using percentages on quantity exported from Italy above. For example for the year 1900 the quantity exported to the US is 331,563,577*0.29=96,153,473 which divided by 159,384,389 provides a percentage equal to 60.32 per cent. ^ In 1898 and 1902 the percentage exported from Italy to the US exceeds the total import into the US. This is because Italian figures refer to the calendar year, while USA figures refer to the fiscal year. Source: Powell (1909)

Table 2: Total Italian export of citrate of lime and total United States imports

Total Italian exports

Total US imports

Year Quantity (pounds)

Value (US $)

Exports to the US (%)*

Quantity (pounds)

Value (US $) Imports from Italy

(%)**

1899 3,142,248 151,295 32.8 1,577,804 157,482 65.3

1900 3,743,448 196,826 37.6 1,944,803 204,243 72.4

1901 3,120,202 147,502 35.3 2,416,658 209,583 45.6

1902 7,517,541 329,965 32.5 3,056,904 293,293 79.9

1903 7,229,647 632,905 38 2,657,110 240,486 100^

*Percentages provided by Powell (1909) ** Percentages estimated using percentages on quantity exported from Italy above. For example for the year 1901 the quantity exported to the US is 3,120,202*0.353=1,101,431 which divided by 2,416,658 provides a percentage equal to 45.57 per cent. ^ In 1903 the percentage exported from Italy to the US exceeds the total import into the US. This is because Italian figures refer to the calendar year, while USA figures refer to the fiscal year

Source: Powell (1909)

Table 3: Distribution of mafia and agricultural production across provinces in 1881-86

Dominant production of:

Province Towns Mafia Citrus Grape

Olive

Wheat Sulphur

Caltanissetta 16 0.437 0.471 1 0.471 1 0.526

Catania 22 0.318 0.478 0.826 0.347 0.782 0.226

Girgenti 17 0.823 0.5 0.611 0.388 1 0.625

Messina 25 0.24 0.608 0.739 0.521 0.478 0.07

Palermo 27 0.296 0.37 0.777 0.444 0.74 0.133

Siracusa 21 0.142 0.35 0.85 0.35 0.75 0

Trapani 15 0.4 0.4 0.8 0.6 0.8 0.066

Total 143 0.357 0.454 0.797 0.441 0.776 0.23

Numbers in the table refer to the share of towns within each province with mafia presence and/or with dominant production of each commodity. Each variable in the table is a binary dummy. Mafia=1 if mafia is perceived to be the most common form of crime in the town and 0 otherwise, as explained in the text. Citrus, Grape, Olive, Wheat and Sulphur are also binary dummies taking on the value of 1 if the commodity is listed by the pretore as one of the key agricultural goods produced in the town, as explained in the text. Source: Damiani (1886)

Table 4: Mafia probit model

Dependent variable: Mafia (in 1880)

(1) (2) (3) (4) (5) (6)

Citrus 1.164** 1.401** 1.488** 1.363** 1.479*** 1.342**

(0.453) (0.567) (0.701) (0.581) (0.540) (0.606)

Grape 0.273 0.620 0.603

(0.371) (0.422) (0.431)

Olive -0.322 -0.451 -0.325

(0.209) (0.291) (0.315)

Wheat 0.455 0.269 0.342

(0.321) (0.402) (0.423)

Sulphur 0.416 -0.195 -0.351

(0.303) (0.387) (0.533)

Population density

-0.0291

(0.0676)

Fractionalization policies

0.831*** 0.970*** 1.100***

(0.246) (0.198) (0.197)

High land fractionalization

-0.128

(0.293)

Large scale plantation

0.962***

(0.173)

Constant -1.47*** -1.31*** -1.461*** -0.891*** -1.144*** -1.311***

(0.411) (0.354) (0.416) (0.203) (0.304) (0.281)

Predicted probability 0.16 0.28 0.31 0.34 0.35 0.39

Area under ROC curve 0.76 0.83 0.85 0.86 0.86 0.88

Provinces All All Without Caltanis-

setta All All All

Province dummies No Yes Yes Yes Yes Yes

Observations 138 138 123 124 113 119

Notes: The estimator is binomial probit in all specifications. Clustered robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

Table 5: Probit regressions with additional controls

Dependent variable:

Mafia (in 1880)

(1) (2) (3) (4) (5)

Citrus 1.327** 1.183** 1.167** 1.472*** 1.318** (0.612) (0.598) (0.547) (0.545) (0.571) Fractionalization policies 1.003*** 0.823*** 0.970*** 1.037*** 0.988*** (0.217) (0.229) (0.258) (0.245) (0.234) Large scale plantation 0.987*** 0.902*** 0.903*** 0.832*** 0.397 (0.232) (0.180) (0.225) (0.256) (0.244) Distance from Palermo (in log) -0.311

(0.295) Distance from Mazzara del Vallo (in log) -0.220* (0.116) Feudal

-0.156

(0.512) Distance from railway station (in log) -0.216*

(0.129) Not care of law (dummy)

1.654**

(0.642) Mistrust in law (dummy)

0.149

(0.528) Trust in law (dummy)

0.584

(0.809) Length of contract (in log)

-0.606***

(0.121) Constant 0.729 -1.671** -1.618** -2.424*** -0.631 (1.528) (0.756) (0.778) (0.906) (0.628)

Predicted probability 0.4 0.35 0.36 0.41 0.34 Area under ROC curve 0.89 0.87 0.87 0.89 0.86 Province dummies Yes Yes Yes Yes Yes

Observations 119 117 112 112 100

Notes: The estimator is binomial probit in all specifications. Clustered robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

Table 6: 2SLS estimates

Panel A

Dependent variable:

Mafia (in 1880)

2nd stage estimates

Estimator IV IVPROBIT

Citrus 0.571*** 2.072*** (0.196) (0.175) Fractionalization policies 0.235*** 0.669* (0.0847) (0.369) Large scale plantation 0.186** 0.459 (0.0841) (0.356) Constant 0.104 -1.838** (0.173) (0.754)

Predicted probability 0.376 0.319 Province dummies Yes Yes

Observations 118 118 R-Squared 0.33

1st stage estimates for Citrus

Excluded Instrument: Mean Free-Frost Period 0.0042*** 0.004***

(0.0011) (0.0013)

Anderson LR Statistics 17.821

Cragg-Donald F-Statistics 17.607

Stock and Yogo 10% critical value 16.38

Partial F-statistics 17.33

Endogeneity Test (p-values) 0.1304

Notes: The estimator is an IV in column 1, and IVPROBIT in column 2. We run a two-stage least square

estimations with Citrus as the endogenous variable and mean free-frost period as the excludable instrument. Robust

standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

Table 7: Falsification tests

Dependent variable:

Grape Olives Wheat Almonds

(1) (2) (3) (4)

Mean Free-Frost Period -0.00117 0.00309 -0.00471 -0.000640

(0.00274) (0.00362) (0.00361) (0.00414)

Constant 1.327 -1.113 2.284* -1.161

(0.850) (1.188) (1.208) (1.362)

Province Dummies Yes Yes Yes Yes Observations 125 142 107 142 Notes: Clustered robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.

Table 8: Mafia intensity and citrus suitability

Dependent variable:

Mafia intensity (Cutrera)

Mafia intensity (IDW)

Oprobit OLS Spatial OLS

OLS Spatial OLS

(1) (2) (3)

(4) (5)

Citrus suitability (rain fed) 2.236*** 1.217** 1.217***

1.167*** 1.167***

(0.807) (0.457) (0.363)

(0.256) (0.242)

Olive suitability (rain fed) 1.237 0.844 0.844

0.626 0.626

(1.301) (1.093) (0.730)

(0.827) (0.688)

Wheat suitability (fain fed) 0.134 -0.00381 -0.00381

-0.654 -0.654

(1.861) (1.532) (1.158)

(0.935) (0.821)

Distance from coast -5.745*** -3.700** -3.700***

-1.329 -1.329

(1.631) (1.223) (1.360)

(0.865) (0.843)

Soil productivity -0.0143 0.0989 0.0989

0.148 0.148

(0.496) (0.330) (0.388)

(0.205) (0.226)

Altitude 7.298*** 4.124*** 4.124***

2.772*** 2.772***

(1.048) (1.036) (0.931)

(0.567) (0.821)

Precipitation -2.026 -1.126 -1.126

0.318 0.318

(3.969) (2.759) (1.197)

(2.334) (1.126)

Inland water scarcity -1.930* -1.075* -1.075***

-0.763*** -0.763***

(1.037) (0.503) (0.334)

(0.226) (0.160)

Land workability 0.692 0.639 0.639***

0.942*** 0.942***

(0.486) (0.356) (0.189)

(0.218) (0.192)

Land nutrient 0.143 -0.0504 -0.0504

-0.0540 -0.0540

(0.476) (0.330) (0.274)

(0.181) (0.146)

Primary net prod. (Rain Fed) -5.099 -3.183 -3.183***

-3.405 -3.405***

(3.642) (3.001) (1.133)

(2.273) (1.037)

Province dummies Yes Yes Yes

Yes Yes

Observations 287 287 287

386 386

R-squared

0.477 0.802

0.702 0.899

Notes: Clustered robust standard errors in parentheses for (1), (2) and (4). Conley (2008) robust spatial standard errors in parentheses for column (3) and (5). Distance cutoff in (3) and (5) equal to 20 km. We tried different cutoff levels and standard errors decreases when we increase the cutoff. *** p<0.01, ** p<0.05, * p<0.1. All indices are scaled on a 0-1 scale.

Table 9: Mafia persistence, 1880-2008

Mafia indicators

Province Mafia (in

1880)

Mafiosi reported in 1999

Mean mafiosi

per capita in 1999

Assets confiscated from the

mafia 1997-2008

Assets confiscated per capita in 1997-2008

Caltanissetta* 0.437 31 0.071 5.368 0.535 Catania 0.318 56 0.051 4.566 0.132 Girgenti (Agrigento)

0.823 9 0.019 5.125 0.447

Messina 0.24 27 0.041 2.384 0.079 Palermo 0.296 30 0.024 70.900 0.855 Siracusa* 0.142 43 0.059 4.045 0.128 Trapani 0.4 5 0.018 17.266 0.679

Total 0.346 201 0.040 19.027 0.377

Notes: Numbers in the first column refer to the share of towns within each province that had mafia in 1880. The second column shows the number of mafiosi murder cases reported in 1999 per province whereas the third column shows the number of reported cases per capita in each province. Columns four to five show the number of assets confiscated from the mafia during 1997-2008 in absolute and per capita terms. See table A1 in the Appendix for an explanation of the variables. *Caltanissetta is merged with Enna and Siracusa with Ragusa as in the Damiani Inquiry. # Population is in thousands

Table 10: Historical mafia presence as a determinant of current mafia activity

Dependent variable: Assets confiscated per capita

1997-2008

(1) (2)

Mafia (in 1880) 0.303** 0.247* (0.140) (0.141) Province dummies No Yes Observations 143 143 R-Squared 0.04 0.19

Notes: The estimator is OLS for both specifications. Robust standard errors in parentheses. *** p<0.01, ** p<0.05,

* p<0.1. Trapani is the excluded province in column 2.

Table 11: Variable descriptions Variables Description Source Mafia Dummy variable for whether the pretori

reports mafia as one of the most common form of crimes The relevant question in the Inquiry is: What is the most important form of crime?

Damiani (1886). Original surveys completed by pretori

Citrus, Grape, Olive, Wheat Dummy Variable proxying the most important crop produced in the town. The question in the Inquiry is: What is the dominant crop in the town?

Damiani (1886). Original surveys completed by mayors excepted for the Province of Caltanissetta for which we use the summary Report (Damiani). Di Vita (1905) is used to check the information in Damiani.

Sulphur mines Dummy variable for whether there is sulphur mine.

Di Vita (1905)

Feudal, Church-Ruled, Crown-Ruled.

Dummy variable coding the political organization within the town before the Unification of Italy.

Di Vita (1905)

Scale of the Plantation Dummy variable for the scale of the plantation. Question: Is the large, medium, or the small scale of the plantation prevalent? To be noticed that there is not a universal criterion to classify the scale. For example the mayor of Caltagirone considers large a plantation extending for more than 2000 hectares and small a plantation extending for less than 20 hectares. On the other hand the mayor of Paterno’ considers large a plantation extending over 50 hectares and small one extending over less than 1 hectare. Most of the times this difference reflects the extent of the town. This difference is important because affects the answer on the fractionalization of land below

Damiani (1886). Original surveys completed by mayors excepted for the Province of Caltanissetta for which we use the summary Report (Damiani).

Fractionalization of Land Dummy variable for the level of fractionalization of land. Question: What is the fractionalization of the land and which factors have affected it? As mentioned above this is affected by the classification of the scale of the plantation above

Damiani (1886). Original surveys completed by mayors excepted for the Province of Caltanissetta for which we use the summary Report (Damiani).

Fractionalization of Policies Dummy variable coded one if policies aimed at increasing peasants’ ownership have been effective in fractionalizing the land. Policies considered are: Enfiteusi, Abolishment of Feudalism, and Privatization of the church’s assets. Question: Which factors have affected the fractionalization?

Damiani (1886). Original surveys completed by mayors excepted for the Province of Caltanissetta for which we use the summary Report (Damiani).

Altitude and Distance from the Railway

Altitude above the sea level (in metres) and distance in Km to the closer railway station.

Di Vita (1905).

Rule of Law Dummy variable for the trust people Damiani (1886). Original

have in the application of law. Question: Do people trust, mistrust, care or fear the law?

Surveys completed by pretori.

Length of Tenancy Contract Question: What is the average length of the tenancy contract? In some cases the mayor reports that the contract lasts for the entire life. In this case we assume an average life expectancy of 60 years and an average duration of the contract of 40 years. We also try with different numbers (20, 30, 50 years) but results do not change.

Damiani (1886). Original survey completed by mayors exception made for the Province of Caltanissetta for which we use the summary Report (Damiani).

Mafia Crimes Number of mafia crimes reported to the police which are punishable according to the 416bis article of the Italian Penal Law.

National Institute for Statistics - ISTAT (1999)

Assets Confiscated Number of assets (i.e. apartments, buildings, etc.) confiscated from the mafia in the town.

Agenzia del Demanio (2008)

Appendix (online publication only)

Table A1: Distribution of lemon trees in Southern Italy in 1898

Region Number of trees

Reggio Calabria 1,232,675

Messina 1,634,231

Palermo 2,488,475

Catania 828,640

Syracuse 460,125

Caltanissetta 8,210

Girgenti (Agrigento) 56,379

Trapani 216,610

Total 6,924,985

Source: Powell (1909)

Table A2: Exports from the harbor of Messina in 1850

Product Quantity Units Lire Description

Silk 1,100 Balle 4,469,850 Oz 645 x Balla

Olive Oil 320,000 Cafissi 2,419,200 Approx 3 Litres x Cafisso

Oranges 500,000 Casse 2,726,325 Approx 240 oranges x cassa.

Lemons 600,000 Casse 3,779,622 Approx 360 lemons x cassa

Lemon Juices 1,000 Barili 503,986 Oz 40 per barile

Salted Lemons 200 Barili 151,200 Oz 6 per barile

Citrus Perfumes 400,000 Libre 2,014,740 Sulphur 90,000 Quintali 302,211 Wheat 50,000 Salme 2,477,790 Flax 20,000 Salme 1,007,937 4 salma = 2 hl

Wine 2,000 Salme 37,800 1/2 salma= 801 hl

Nuts 4,000 Salme 655,169 4 salma = 3 hl

Almond 20,000 Cantaia 1,763,370 Oz 7 x cantaio

Pistachio 200 Cantaia 30,240 Oz 12 x Cantaio

Walnuts 2,000 Salme 50,387 4 salma = 3 hl

Liquorice 16,000 Cantaia 680,400 Oz 9 per cantaio

Sardines 4,000 Barili 151,162 Oz 2 per Barile

Carob 4,000 Sacchi 90,720 24 sacchi = 90Kg

Wool 2,000 Cantaia 453,600 6 cantaio = 80Kg

Linen 7,000 Quintali 264,600 Oz 3 x quintali

Cotton 4,000 Quintali 30,240

Source: Battaglia (2003)

Table A3: Descriptive statistics for political organization across provinces before the Italian unification in 1860

Political organization:

Province Towns Feudal Crown-ruled Church-ruled Frac. policies

Caltanissetta 12 0.833 0.167 0 0.714

Catania 31 0.452 0.29 0 0.625

Girgenti 24 0.583 0.125 0 0.714

Messina 28 0.357 0.28 0.214 0.285

Palermo 30 0.566 0.267 0.133 0.346

Siracusa 22 0.818 0.181 0 0.5

Trapani 15 0.666 0.333 0 0.285

Total 162 0.574 0.216 0.0617 0.413

Numbers in the table refer to the share of towns within each province that were characterized by feudal organization, was crown- or church-ruled, or had a substantial degree of fractionalization polices. Each of the variables of Political organization is a binary dummy. Source: Damiani (1886)

Table A4: Land fractionalization and scale of plantations by province in 1881-86

Fractionalization of ownership

Scale of plantation

Province Towns High Low Large Small

Caltanissetta 12 0.545 0.454 0.333 0.5

Catania 23 0.409 0.227 0.26 0.695

Girgenti 14 0.461 0.461 0.357 0.461

Messina 20 0.277 0.444 0.35 0.789

Palermo 26 0.52 0.16 0.461 0.576

Siracusa 20 0.35 0.5 0.35 0.666

Trapani 14 0.333 0.333 0.285 0.923

Total 129 0.413 0.347 0.348 0.661

Numbers in the table refer to the share of towns within each province that were characterized by high or low fractionalization in land ownership and large or small plantations. Source: Damiani (1886)

Table A5: Pairwise correlations

Mafia Citrus Wheat Olive Grape Sul-phur

Feudal Frac. policy

Pop. density

High frac.

Large scale plant.

Mafia 1

Citrus 0.3863 1

Wheat 0.0946 -0.239 1

Olive 0.0014 0.2289 -0.185 1

Grape 0.068 0.0424 -0.015 0.1391 1

Sulphur 0.1914 -0.0124 0.2903 -0.036 -0.058 1

Feudal 0.0556 -0.1212 0.0598 -0.011 -0.132 0.0519 1

Frac. policies

0.2637 0.0713 0.2214 -0.056 0.0552 0.1771 0.0531 1

Pop. density -0.05 0.1145 -0.341 -0.039 0.0492 -0.182 -0.237 -0.057 1

High land fract.

0.0413 0.0491 -0.104 0.0946 -0.065 -0.048 -0.038 0.073 0.1596 1

Large scale plant.

0.2465 0.1623 0.1054 -0.020 0.1406 0.0299 0.0642 -0.15 -0.2317 -0.1886 1

Table A6: Descriptive statistics for robustness analysis

Obs Mean Std. Dev. Min Max

Mafia intensity (Cutrera, 1900) 289 1.435986 1.128953 0 3

Mafia intensity (IDW interpolation) 391 1.264872 .9225032 0 2.956

Citrus suitability 386 .2374887 .1436314 0 1

Olive suitability 391 .65047 .2441069 0 1

Wheat suitability 391 .5571812 .2065363 0 1

Figure A1: Optimal mafia effort m* at varying level of profits π*

Note: The initial example above (giving the curves m* and ω*) assumes the following parameter levels: d=0.4, θ=0.5,

π*=8, A=10. In the second example, we alter only the level of profits to π*=32 which yields the new curves m** and

ω**. The equilibrium level of mafia activity thereby increases from m*=0.09 to m**=0.236.

10.750.50.250

0.625

0.5

0.375

0.25

0.125

0

om ega

m

om ega

m

Profits

increase

m*

m**

ω*=0.526 ω**=0.297

Figure A2: Variables in the original Inquiry Damiani (1886) related to mafia

Note: Figure A2 shows a picture of the second part of the table from which we got data on mafia. The first row reports variables. Starting from the left, the first

variable is Scostumatezza (lewdness). There are three possible causes among which the prefect can choose. The first one is Adulterio (adultery), the second Incesto

(incest), the third is Nascite Illegali (illegally born child), and the last one is Varie (various). The second variable relates to the religiousness of people in the town,

the third one relates to the clergy (corrupt or exemplary), and the fourth variable regards perjury. The last variable is Vagabondaggio accattonaggio (vagrancy and

begging). The fifth variable is the one we use to get information on mafia. The variable is labelled Reati (crimes) and then it is asked what the most common crimes

are and the extent of these crimes. Most of the time, pretori only answered providing information on the sort of crime committed. The most common forms of

crime were rustling, mafia, bloody crimes, and bloody crimes for passion. In addition poverty was described as the most common cause of crime.

Figure A3: Inchiesta Agricola (Agricultural Inquiry) in Damiani (1886)

Note: Figure A3 shows the first page of the Agricultural Inquiry for the city of Cefalu. The column on the left reports the question (Quesiti). The column on the

right reports the answer of the mayor. For example the first question asks “What is the surface area of the city”, “In how many areas the territory can be divided?”

and “What is the extent of each area?” The mayor answers that the surface land of Cefalu is 52,943,492 sq.mt. and the territory is divided in three zones: 1) plain; 2)

hills; 3) mountains. The first zone extends for almost 2,500,000 sq.mt.; the second for almost 20,000,000 sq.mt.; and the third for almost 30,443,492 sq.mt. The

second question relates to the physical and chemical characteristics of the territory. This first part of the inquiry which is titled the Condition of Agriculture also

reports information on the kind of crops produced, the sort of manufactures developed in the city, and so on. The second part of the Inquiry relates to the

relationship between peasants and landlords, while the third part regards the moral conditions of peasants.

Figure A4: Cutrera’s map

Figure A5: The current distribution of crops