یکشنبه ۰۸ مهر ۰۳ | ۱۱:۵۵ ۷ بازديد
Figure 2
Drought intensity in spring 1893. Ratio of the rainfall in spring 1893 to long-run average spring rainfall.
Figure 3
Mafia intensity in 1900 according to the Police inspector Cutrera. Source: Cutrera (1900).
having a major Mafia presence in the Siracusa province. We believe that this is a reliable source
for the Mafia intensity, and it is widely acknowledged in the historical literature. The geographic
distribution of the Mafia intensity according to Cutrera (1900) is depicted in Figure 3.
To control for the presence of the Mafia before the Fasci movement, we use the information
reported in the 1880–85 parliamentary inquest on the Italian agriculture (Damiani, 1885). Damiani
requested information about the presence of the Mafia from local state officials in each of the 162
Sicilian judicial districts. We follow Buonanno et al. (2015) and classify all the municipalities in
Downloaded from https://academic.oup.com/restud/article-abstract/87/2/537/5364272 by MIT Libraries user on 29 May 2020
Figure 4
Spatial distribution of the Mafia presence according to its intensity as reported in Damiani (1885).
the judicial district according to the intensity of the Mafia. In Figure 4 we depict the resulting
variable, Mafia1885.
As for our controls, we collected data about the agricultural use of land and rents at municipality
level from Sicily’s Bourbon Cadastre compiled by Mortillaro (1854). The agricultural use of land
is based on the riveli, declarations made by the landowners to the Royal commission in 1815 and
updated in the 1830s. This dataset is particularly relevant because it pertains to a period before
Italian unification and the Mafia. From this source we obtain the average agricultural rent, and
for municipalities with more than 3,000 residents, urban rent per hectare. We also compute from
the same source the total area of the municipality, the share of agricultural land, and the share of
the total area devoted to cereal cultivation, citrus groves, vineyard, and olive groves.
From the first Italian census in 1861 we gathered information on total population as well as on
the existence of agro-towns (rural centres) where peasants and farmers were concentrated. These
towns are a peculiar characteristic of the Sicilian countryside, often linked to a concentrated
landownership structure, which induced landless peasants to gather in rural town centres where
they could get seasonal or other work in large estates specializing in cereals. The presence of
such towns is thus both an indication of the pattern of landownership and, as emphasized by
Hobsbawm (1971), a potential direct determinant of peasant organizations, since the presence of
so many peasants agglomerated in agro-towns “gave them the opportunity to discuss grievances,
formulate unified strategies and act collectively” (Kaplan, 1977, p. 5). We define agro-towns as
municipalities that have a centre with an agglomerated population of at least 4,000 residents and
with more than 40% of the agricultural land devoted to cereal cultivation.
As highlighted in the historical background, another important determinant of the development
of the Sicilian Mafia has been the presence of sulphur mines. We rely on data reported in Parodi
(1873) for the average production of sulphur in each municipality between 1868 and 1870.
We use further geographical variables to control for the distance from the capital, Palermo,
and from the closest port using the postal distance as reported by Lo Jacono (1856) which takes
into account the availability of roads. A dummy variable for direct access to one of the postal
roads is coded from the detailed maps compiled by Cary (1799) and digitized by Buonanno et al.
(2015). Information on the maximum and average altitude and the altitude of the town centre is
Downloaded from https://academic.oup.com/restud/article-abstract/87/2/537/5364272 by MIT Libraries user on 29 May 2020
obtained from the 1929 Agrarian Cadastre. Finally, the average temperature is from the website
climate-data.org.
Data on political competition in 1865 and 1909 are obtained from the database provided by
Corbetta and Piretti (2009). The parliamentary elections in 1865 and 1909 were held under a first-
past-the-post system in fifty-five electoral districts in Sicily. For each municipality, we compute
the Herfindahl concentration index (or HHI as an abbreviation for Hirschman-Herfindahl index)
by looking at the share of votes of each candidate. Using data from the same 1909 parliamentary
elections we compute the share of votes of socialist candidates as well. We collected voting data for
the lower chamber of the Italian parliament between 1948 and 1992 from the Election Historical
Archive of the Interior Ministry website. These elections were held under a proportional system
with a single national constituency and we compute the HHI from vote shares of all the parties in
each municipality. As an alternative measure of the presence of socialist organizations, we also
use data on peasant league membership reported in Lorenzoni (1910b).
We measure human capital, our main ******* for the overall level of economic development,
using data on literacy and high school completion rate for the population above the age of six
from the censuses for the years 1911, 1921, 1931, 1961, 1971, and 1981. 18
As a ******* for the provision of public goods, we constructed from primary sources average
infant mortality rate for 1869–70 and 1908–9 for each municipality. Specifically, we counted the
number of deaths of residents below the age of one from the death registries of each municipality
for 1869, 1870, 1908, and 1909 (when one of the registries was not available, we used data from
the subsequent year). We then collected data on the number of births for the corresponding years
and computed the infant mortality rate as the average ratio of deaths below the age of one over the
number of births for the years 1869 and 1870, and 1908 and 1909 by municipality of residence.19
For infant mortality rates in 1969–70 and 1982 we rely on data provided by ISTAT. 20 Information
on the quantity of drinking water in 1885 is from Direzione Generale della Statistica (1886), and
data on aqueducts in 1961, 1971, and 1981 are from the Population and Housing Censuses. As a
******* for local state capacity, we also look at the number of notaries per 1,000 inhabitants at the
municipality level, which are constructed from the data of the Finance Ministry on taxpayers for
1871 and 1924 (Ministero delle Finanze, 1872, 1924).
We computed a measure of per capita development expenditure (in logs) from the 1884, 1912,
and 1957 municipality budgets. This measure aggregates all spending under the discretion of the
local administration directed at economic development or improvement of local infrastructure,
education, security, and justice. We collected these data from official government publications
(Direzione Generale della Statistica, 1887; Direzione Generale della Statistica e del Lavoro,
1914; Direzione Generale dell’Amministrazione Civile, 1958).
We finally estimate state presence before 1893 using two separate measures: the number
of soldiers stationed in Sicily used for local policing divided by population and the efficiency
of civil courts. We construct estimates for the first variable using data from the Parliamentary
Commission of 1875, and in particular a map from the report of the commanding general
18. High school data are not available for the period 1911–31.
19. For four municipalities (Caltanissetta, Girgenti, Naro, and Noto), the mortality data are available for one year
only in the period 1869–70 and the mortality rate is computed accordingly. In two cases (Messina and Catania), the
mortality rate for 1908–9 has been based on the data for 1908 only. The data for the number of births by municipality are
from Ministero Agricoltura, Industria e Commercio (1872) and Somogyi (1979).
20. The data for the years 1969–70 report the deaths by municipality of the event instead of municipality of
residence. There is therefore a mismatch between the numerator (deaths by municipality) and the denominator (births by
municipality of residence) which likely increases the mortality rate for municipalities with hospitals and health centres.
Downloaded from https://academic.oup.com/restud/article-abstract/87/2/537/5364272 by MIT Libraries user on 29 May 2020
(Carbone and Grispo, 1969). 21 The efficiency of civil courts in 1875 is measured as
the share of civil and commercial cases receiving a final sentence within the year
(Ministero di Grazia e Giustizia e dei Culti, 1877).
Table 1 provides descriptive statistics for all of the variables described in this section.
4. THE ORIGINS OF THE 19TH-CENTURY EXPANSION OF THE MAFIA
This section empirically investigates the effects of the drought of 1893 and the Peasant Fasci
movement on the spread of the Sicilian Mafia. It provides evidence that the drought of 1893 had
a major impact on the presence of Peasant Fasci and the spread of the Sicilian Mafia.
4.1. The drought and agricultural production
We start by confirming that the drought of spring 1893 had a sizable negative impact on
agricultural production. We use data from Direzione Generale dell’Agricoltura (1896–96) to
measure agricultural production for eighteen crops in twenty-four districts in 1893. To control
for potential differences in agricultural productivity across the districts, we divide production in
1893 by the average of the production of the crop in that district over the years 1885–92 and
1894–95. We combine this with our data on relative rainfall in the spring of 1893 to estimate the
following relationship:
relative productionc,d = κ ·relative rain 1893
d +η ·Kc ·relative rain 1893
d +X′
c,d βprod +ψc +εprod
i .
(1)
Here, relative rain1893
d is relative rainfall in the district, capturing the (inverse of the) severity of
the drought of 1893, while Kc is a measure of the relative importance of spring rainfall for crop
c.22
The covariates include a full set of crop dummies (ψc), the main effects of various different
rainfall measures, information on the crop’s productivity in the district (1885–95) and its cultivated
area in 1893, and province or district dummies in some specifications.
The results are reported in Table 2.23 We start in columns 1 and 2 without the interaction terms,
thus focusing on the main effect of relative rainfall in the spring of 1893. The results confirm
that districts suffering worse drought conditions (lower relative rainfall) in the spring of 1893
had lower agricultural production. Rainfall in the winter of 1892 is also positive and significant
21. There were 7,200 soldiers (equivalent to 22.5 battalions) stationed in Sicily used for local policing at that time.
The map provides information on where each battalion, company, platoon and squad was located, but we do not know the
exact size of each unit. We subtract all soldiers in companies, platoons and squads from the total number of soldiers and
then divide the remaining soldiers among the battalions. The number of soldiers in a municipality is then estimated as the
sum of the sizes of battalions, companies, platoons, and squad in that municipality. In this exercise, we use estimates of
the number of soldiers per squad and platoon consistent with information in Encyclopaedia Britannica: ten soldiers per
squad and thirty soldiers per platoon. We show in the Supplementary Appendix that our results are robust to alternative
assumptions about the sizes of squads and platoons.
22. The Kc coefficient is defined by the Food and Agriculture Organization (FAO) as the ratio between the
evapotranspiration (sum of evaporation from the land surface and transpiration from plants) of the crop in question
relative to a reference crop assumed to be a uniform grass field. Kc varies during the different stages of crop growth,
and we computed the spring Kc by averaging over the spring period using the crop water information provided by the
FAO. See Brouwer and Heibloem (1986) and the FAO website (www.fao.org/land-water/databases-and-software/crop-
information/en/) for further information.
23. We report bootstrapped standard errors allowing for two-way clustering conditional on the district and the crop;
details on the two-way bootstrapping procedure are provided in the next subsection.
Downloaded from https://academic.oup.com/restud/article-abstract/87/2/537/5364272 by MIT Libraries user on 29 May 2020
TABLE 1
Descriptive statistics
Variable Obs Mean SD Min Max Variable Obs Mean SD Min Max
Main variables Economic, Public Goods and State Capacity variables
Mafia 1900 273 1.43 1.15 0 3 Literacy rate 1911 333 0.37 0.10 0.11 0.73
Peasant Fasci 333 0.31 0.46 0 1 Literacy rate 1921 333 0.48 0.09 0.23 0.80
Relative Rainfall 1893 297 0.64 0.28 0.06 1.28 Literacy rate 1931 333 0.59 0.08 0.37 0.79
Literacy rate 1961 333 0.83 0.04 0.69 0.93
Determinants of Fasci Literacy rate 1971 333 0.88 0.04 0.74 0.99
Peasant Fasci before March 1893 333 0.02 0.12 0 1 Literacy rate 1981 333 0.93 0.03 0.81 0.99
Rural centre in 1861 333 0.37 0.48 0 1 High school rate 1961 333 0.03 0.02 0.01 0.10
Rural rent per hectare in 1853 333 6.85 4.75 1.17 35.15 High school rate 1971 333 0.06 0.03 0.02 0.22
Urban rent per hectare in 1853 333 1.56 4.30 0 69.99 High school rate 1981 333 0.09 0.04 0.03 0.30
Grains (share in 1830s) 333 0.50 0.23 0.015 0.99 Infant mortality 1869-70 259 0.20 0.07 0.01 0.67
Share cultivated land in 1853 333 0.97 0.08 0.17 1 Infant mortality 1909-10 277 0.16 0.04 0.04 0.32
Infant mortality 1969-70 333 0.03 0.02 0 0.11
Determinants of Mafia Infant mortality 1982 333 0.02 0.02 0 0.11
Sulphur production 1868-70 333 5.59 23.50 0 210 Drink water quantity in 1885 333 1.69 0.51 1 3
Citrus groves (share in 1830s) 333 0.01 0.02 0 0.16 Aqueduct coverage 1961 333 0.66 0.24 0.01 0.99
Mafia 1885 333 0.57 1.01 0 3 Aqueduct coverage 1971 330 0.77 0.12 0.11 1
Olives (share in 1830s) 333 0.04 0.06 0 0.46 Aqueduct coverage 1981 333 0.67 0.14 0.01 0.99
Vineyards (share in 1830s) 333 0.10 0.12 0 0.74 Development Expenditure 1884 333 0.82 0.85 −2.15 3.28
Development Expenditure 1912 333 −1.58 2.46 −10.03 5.23
Geographic controls Development Expenditure 1957 333 0.20 0.93 −2.76 3.56
Population in 1861 333 8.37 0.95 5.88 12.19 Notaries per 1000 inhabitants in 1871 333 0.01 0.05 0 0.52
Area 333 8.194 1.25 4.41 11.19 Notaries per 1000 inhabitants in 1924 333 0.15 0.16 0 0.98
Altitude of the town centre 333 411.40 276.60 3 1265 Soldiers per 1000 inhabitants in 1875 333 2.11 4.30 0 35.86
Maximum altitude 333 944.20 591.70 48.00 3274 Efficiency of Civil courts in 1875 333 0.40 0.13 0 0.72
Average altitude 333 392.10 274.90 10 1627
Roads in 1799 333 0.51 0.50 0 1 Political competition
Distance from Palermo 333 109.02 58.91 0 229 HHI 1865 289 0.77 0.20 0.45 1
Distance from closest port 333 32.49 20.13 0 90 HHI 1909 317 0.78 0.22 0.34 1
Average temperature 333 15.93 1.54 11.00 18.45 HHI 1963 333 0.31 0.07 0.17 0.59
Long run average rainfall (1881-1941) 307 156.30 29.18 70.15 253.10 HHI 1972 333 0.31 0.07 0.18 0.65
Long run variance of relative rainfall 307 0.21 0.07 0.12 0.47 HHI 1983 333 0.30 0.07 0.16 0.57
Notes: The descriptive statistics include the number of observations (Obs), the average (Mean), the standard deviations (SD), the minimum (Min) and the maximum value (Max) for the
entire sample of municipalities. See text for variable definitions and sources.
Downloaded from https://academic.oup.com/restud/article-abstract/87/2/537/5364272 by MIT Libraries user on 29 May 2020
TABLE 2
The impact of relative rainfall on agriculture production in 1893. District-level data
Dependent variable: change in crop output per hectare in 1893
(1) (2) (3) (4) (5) (6) (7) (8)
Relative Spring Rainfall 1893 × Kc spring 1.44 1.45 1.46 1.45 1.71 1.71
coefficient (0.73) (0.70) (0.69) (0.66) (0.99) (0.95)
Relative Spring Rainfall 1893 0.40 0.42 −0.69 −0.67 −1.00 −1.19
(0.13) (0.12) (0.58) (0.56) (1.23) (1.33)
Relative Fall Rainfall 1892 −0.02 −0.02 −0.03 −0.03 −0.43 −1.12
(0.16) (0.15) (0.16) (0.15) (1.11) (1.43)
Relative Winter Rainfall 1892-3 0.31 0.28 0.30 0.28 0.17 0.31
(0.16) (0.16) (0.16) (0.16) (2.31) (2.37)
Relative Summer Rainfall 1893 −0.08 −0.08 −0.08 −0.07 −0.02 0.01
(0.05) (0.05) (0.05) (0.05) (0.28) (0.46)
Relative Fall Rainfall 1892 × Kc spring 0.91 0.92
coefficient (1.17) (1.14)
Relative Winter Rainfall 1892-3 × Kc spring −0.18 −0.16
coefficient (1.19) (1.16)
Relative Summer Rainfall 1893 × Kc spring −0.04 −0.08
coefficient (0.49) (0.47)
Crop fixed effect
Crop-district specific average output per ha
1885-95
Crop-district specific cultivated area in 1893
Province fixed effects
District fixed effects
Observations 303 303 303 303 303 303 303 303
R2 0.19 0.21 0.20 0.23 0.25 0.28 0.25 0.28
Notes: OLS estimates of the impact of relative rainfall in the spring of 1893 on the production of several crops. The
dependent variable is the change in output per hectare in 1893 with respect to the average for the years 1885–95 (excluding
1893) for 18 crops at district level (24 districts). The crops are barley, beans, broad beans, chestnuts, corn, flax, hemp,
hay, lemons, oat, olive oil, oranges, other citrus, potatoes, rice, rye, wheat, wine. Relative rainfall is measured at weather
station level and interpolated at municipality level using the inverse of the distances as weights with a cutoff of 30 km.
The district-level relative rainfall is the average of municipality-level data. In column 1 we include relative rainfall in the
spring of 1893, relative rainfall in the other seasons and crop-specific fixed effects. In column 2, we add the crop-district
specific average output per hectare and the area devoted to the specific crop in the districts in 1893 in logs as additional
controls. In column 3, we replicate the specification of column 1 and add our variable of interest which is relative rainfall
in the spring of 1893 multiplied by the crop-specific spring-rainfall coefficient (Kc). The latter is the evapotranspiration
of the crop in question relative to a uniform grass field which is the reference crop. It varies during the different stages
of crop growth and we computed the spring K c by averaging over the spring period using the crop water information
provided by the FAO. See Brouwer and Heibloem (1986) and the FAO website (www.fao.org/land-water/databases-and-
software/crop-information/en/) for further information. The crops with high spring Kc are essentially grains: rice with
Kc = 1; rye, barley, wheat and oat with Kc = 0.9. The crops with low spring Kc are wine (0.4), chestnuts (0.5), and citrus
and olive oil (0.7). In column 4, we replicate the specification of column 2 including relative rainfall in the spring of 1893
multiplied by the crop-specific spring-rainfall coefficient (Kc). In column 5 we add province fixed effects. In column 6,
we replicate the specification of column 5 including district fixed effects (and therefore drop the seasonal relative rainfalls
which are constant across crops within district). In column 7 as a falsification test we include the interactions between the
the crop-specific spring-rainfall coefficient (Kc) and the rainfall in all the other seasons. Finally, in column 8 we replicate
the specification of column 7 with district fixed effects. We report bootstrapped standard errors allowing for two-way
clustering conditional on the district and the crop.
in column 1, reflecting the fact that crops sown in late fall, mainly grains in Sicily, benefit from
winter rainfall as well. 24
The remaining columns of the table turn to our main focus, the interaction between relative
rainfall in the spring of 1893 and Kc, represented by η in equation (1). The estimate of η in
column 3 is 1.44 (standard error = 0.73) and implies that a more severe drought in the district
24. See Palmieri (1883) and http://www.fao.org/land-water/databases-and-software/crop-information/wheat/en/.
Downloaded from https://academic.oup.com/restud/article-abstract/87/2/537/5364272 by MIT Libraries user on 29 May 2020
leads to a bigger drop in output for a crop that is more dependent on spring rainfall such as
wheat or barley (in specifications with interactions the main effects of the rainfall variables are
evaluated at Kc = 0). The relationship becomes a little more precise in columns 4 and 5 when we
control for crop-specific output and cultivation area, and add province fixed effects. Our preferred
specification is reported in column 6 and includes a full set of district fixed effects (and thus drops
the main effects of various rainfall measures). In this case, the coefficient estimate of η is 1.45
(standard error = 0.66). This implies that a district suffering a very severe drought, corresponding
to the 25th percentile of the distribution of relative rainfall in spring 1893, experiences a 6%
decline in the production of wine (with Kc = 0.4) and a 32% decline in the production of olive
oil (Kc = 0.7), but a much larger decline for wheat (Kc = 0.9), of almost 85%.
Columns 7 and 8 report specifications that can be interpreted as falsification checks where
we in addition include interactions between our measure of the importance of spring rainfall, Kc,
and relative rainfall during other seasons. Consistent with our interpretation, we find that these
interactions are smaller and insignificant, while the coefficient of the interaction with relative
spring rainfall remains positive and significant at 10%.
Overall, this evidence verifies that there was a large decline in agricultural production in parts
of Sicily most adversely affected by the drought of 1893 and that this was much more pronounced
for crops such as wheat that depend most heavily on spring rainfall.
4.2. First stage
Our first stage is given by the following cross-sectional relationship linking the presence of the
Peasant Fasci to the drought of 1893:
Fascii = γ Fasci ·relative rain 1893
i +X′
i βFasci +εFasci
i , (2)
where Fascii denotes our dummy variable designating the presence of the Peasant Fasci in
municipality i, relative rain 1893
i is the relative rainfall in the spring of 1893, now for municipality
i, Xi is a vector of covariates which are discussed in greater detail below, and εFasci
i is a random
error term, capturing all omitted factors. Equation (2) will be our first stage when estimating the
impact of the Peasant Fasci on the development of the Mafia.
Unless stated otherwise, we report bootstrapped standard errors allowing for two-way
clustering conditional on the district in which the municipality is located and the closest weather
station to the municipality (Cameron et al., 2008, 2011). These standard errors take into account
the fact that rainfall in a municipality, as well as many of our other variables, might exhibit
significant correlation with neighbouring municipalities. 25 Moreover, as described in the previous
section, and further discussed in our robustness checks, the fact that rainfall data for many
municipalities are interpolated from neighbouring weather stations creates additional correlation
across observations; the two-way clustering is a simple method to remove this correlation.
As we will see below, maximum likelihood estimation taking the exact structure of the error
term resulting from interpolation leads to standard errors that are very similar to the two-way
clustered errors, while other corrections for spatial correlation we explore below turn out to be
less conservative. 26
25. We use thirty weather stations across twenty-four districts. Nine out of thirty-nine weather stations we have
available do not have complete data for all three of the months of March, April, and May 1893, which we use to construct
our (relative) rainfall measure.
26. For example, in Table A1 in the Supplementary Appendix we report Conley’s spatially-corrected standard errors
as well as non-corrected (white noise) standard errors. Both are smaller than our two-way bootstrapped errors. Finally,
Downloaded from https://academic.oup.com/restud/article-abstract/87/2/537/5364272 by MIT Libraries user on 29 May 2020
TABLE 3
The impact of relative rainfall 1893 on Peasant Fasci
Dependent variable: Peasant Fasci
(1) (2) (3) (4)
Panel A: without province fixed effects
Relative Rainfall 1893 −1.00 −0.94 −0.93 −0.78
(0.14) (0.13) (0.14) (0.20)
R2 0.36 0.41 0.44 0.46
Panel B: with province fixed effects
Relative Rainfall 1893 −0.76 −0.81 −0.74 −0.72
(0.21) (0.24) (0.25) (0.29)
R2 0.39 0.44 0.48 0.49
Determinants of Fasci
Determinants of Mafia
Geographic controls
Observations 245 245 245 245
Notes: OLS estimates of the impact of relative rainfall in the spring of 1893 on the emergence of Peasant Fasci
organizations. Relative rainfall is measured at weather station level and interpolated at municipality level using the
inverse of the distances as weights with a cutoff of 30 km. The dependent variable is a dummy indicating the presence in
the municipality of a Peasant Fasci organization. Panel A does not include province fixed effects which are included in
all the specifications of Panel B. The specifications in column 1 include only relative rainfall in 1893. The specifications
in column 2 also include other determinants of the presence of the Peasant Fasci (a dummy indicating whether a Peasant
Fasci was present before March 1893, a dummy for the municipality being an agro-town, the levels of rural rents and
urban rents in 1853, the share of total cultivated land, and the share of land devoted to grains). The specifications in
column 3 also include various determinants of the presence of the Mafia (sulphur production in 1868–70, the share of
land devoted to citrus groves, vineyards and olive trees, and a measure of the presence of the Mafia in 1885). Finally,
in column 4 we include a range of geographic controls (log population in 1861, log area of the municipality, elevation
of the town centre, maximum altitude, average altitude, distance to Palermo, distance to the closest port, the access to
a postal road, average temperature, average rainfall, and variance of relative rainfall). We report bootstrapped standard
errors allowing for two-way clustering conditional on the district in which the municipality is located and the closest
weather station to the municipality.
Estimates of equation (2) on our sample of 245 municipalities are reported in Table 3. 27 In
column 1, Panel A we show the raw correlation between our Peasant Fasci variable and relative
rainfall (without controlling for any covariates). The coefficient estimate is −1.00 and is significant
at 1%.
Column 1, Panel B adds province fixed effects to this specification so that the comparison
is between municipalities in the same province (as noted above, we have seven provinces in the
data). The relationship is quite similar and has a somewhat smaller magnitude in the presence of
province fixed effects; −0.76 (with a two-way clustered standard error of 0.21).
The remaining columns add successively more of the covariates we use throughout the article
— with and without province fixed effects in the two panels. In column 2, we include various
determinants of the presence of the Peasant Fasci, which are: a dummy for the presence of a
in Supplementary Tables A2 and A3 we estimate equations (2) and (3) at the district level, which can be interpreted as a
very conservative correction for district-level correlation of error; the results are similar in this case (though because we
only have twenty-three districts and thirty covariates and province dummies, in this case we include different subsets of
the covariates separately).
27. As explained in Section 3, in our source for the Mafia presence in 1900 (Cutrera, 1900) sixty-eight municipalities
were left unclassified. Furthermore, thirty-six municipalities do not have an active weather station within a 30 km radius
in spring 1893 and their rainfall data are missing. These missing data reduce our sample from 333 municipalities to 245.
Downloaded from https://academic.oup.com/restud/article-abstract/87/2/537/5364272 by MIT Libraries user on 29 May 2020
Peasant Fasci before spring 1893, a dummy for the municipality being a rural centre (agro-town),
the levels of rural rents and urban rents, the share of total cultivated land, and the share of land
devoted to grains. These variables are expected to affect the presence of the Fasci, which were
more likely to organize in places where grain production was important. The inclusion of these
covariates has very little effect on the relationship between our drought variable and the presence
of the Peasant Fasci, and the estimates of γ Fasci are now −0.94 and −0.81, without and with
province fixed effects, respectively.
Column 3 in addition includes various determinants of the presence of the Mafia: sulphur
production in 1868–70 (emphasized by Del Monte and Pennacchio (2012) and Buonanno et al.
(2015)), citrus groves in 1830s (emphasized by Dimico et al. (2017)), olives and vineyards
(discussed by Bandiera (2003)), and Damiani’s measure of the strength of the Mafia in 1885.
These variables also have very little effect on the coefficient estimates of interest, which now
stand at −0.93 and −0.74 without or with province fixed effects. Column 4 adds a range of
geographic controls, in particular, log population in 1861, log area of the municipality, maximum
and average altitude and the elevation of the town
dfs3434