tesis de magÍstereconomia.uc.cl/docs/tesis_fgonzalez.pdf · tesis de magÍster instituto de...
TRANSCRIPT
D O C U M E N T O D E T R A B A J O
Instituto de EconomíaTESIS d
e MA
GÍSTER
I N S T I T U T O D E E C O N O M Í A
w w w . e c o n o m i a . p u c . c l
-�� ������ �� .� ����� � ������
"���' ����� �� � �/� �� �������
�� � ��������
����
PONTIFICIA UNIVERSIDAD CATOLICA DE CHILE I N S T I T U T O D E E C O N O M I A MAGISTER EN ECONOMIA
Land Reform and Government Support: Voting Incentives in the Countryside
Felipe A. González Ramírez
Comisión
José Díaz Francisco Gallego
Matias Tapia Rolf Luders Gert Wagner
Santiago, julio 2010
LAND REFORM AND GOVERNMENT SUPPORT: VOTING
INCENTIVES IN THE COUNTRYSIDE∗
Felipe Gonzalez
This paper studies the effects of land reform on political support for the incumbent party.
Using agricultural and housing census data at the county level two major findings are
presented. First, using different estimation techniques I found that incumbent support
increases in about 5% in counties with land reform. Second, agricultural workers seem
to be the main group changing its voting patterns in these counties. I discuss several
mechanisms that could be behind these results. I also use land redistribution that made
the Church with its own land among agricultural workers as a falsification exercise and
robustness check. This exercise provides further support to my main results.
I am here to fulfill my promises, to stand strong by my beliefs and to never weaken my position (...)
I am here because I wish too see the fall in the concentration of land, so that farmers can become
landowners in order to produce their own income and, thus, have a fair wage.
Eduardo Frei Montalva, first speech as President of Chile (November, 1964).
1 Introduction
Land reform was an important economic policy during the sixties in Latin America. In
1961, during the Punta del Este conference and under the general consensus of all Latin
American governments, the Alliance for Progress was born. A main objective of this Alliance
was to make a deep transformation of unfair agrarian structures (Huerta 1989, p.14). Chile
was not the exception: its high land concentration and limited ability to feed the growing
population with its agricultural production lead to the general agreement that an agrarian
reform was needed (Tello, 1965). Thus, in 1962 an agrarian reform began under the right
∗August, 2010. Thesis written as a Master student at the Economic History and Cliometrics Lab (EH Clio
Lab, Pontificia Universidad Catolica de Chile, Department of Economics). I would like to thank Francisco
Gallego, Gert Wagner, Jose Dıaz, Matıas Tapia, Rolf Luders, Claudio Ferraz, Guillermo Marshall, Carlos
Alvarado, and Ignacio Cuesta for useful comments and suggestions. Any comment to the author’s email
address [email protected]
wing government of Jorge Alessandri (1958 – 1964) and then continued under the centre
government of Eduardo Frei (1964 – 1970).
This happened after the introduction of the secret ballot (1958), which prevented landown-
ers to buy the votes of agricultural workers (Robinson and Baland, 2008). In the sixties,
therefore, a new class of voter became important from the perspective of political coali-
tions: agricultural workers. This is relevant because despite the general view of Chile as a
copper producer, agriculture is also an important economic activity and rural laborers rep-
resented a large share of the population (more than 65% of the labor force in counties like
Freire and Calbuco, 1970). If land reform affects agricultural workers, or some variable that
they take into consideration when they evaluate different political alternatives, they might
change their voting patterns in response. This paper precisely analyzes this and examines
whether land reform during the sixties affected support for the incumbent party at the 1970
presidential election. It also examines different mechanisms to explain why this could have
happened. My hypothesis is that land reform increased incumbent support and agricultural
workers are the voters who explain this. This relies on the presumption that land reform
is beneficial for them. To test this I use disaggregated data at the county (municipality)
level, the smallest administrative unit.
The study of how voters react to government policies is vast, and several channels
through which a government policy might affect political preferences of people have been
proposed. The two most common examples of how voters could react to policies are, first,
to consider voter’s reactions to macroeconomic conditions like the rate of unemployment
and income growth (Stigler 1973, Kramer 1971, Fair 1978, see Hibbs 2006 for a review and
Cerda and Vergara (2007) for the Chilean case); and second, to consider voter’s reactions to
government expenditures, transfers, or redistributive policies in general (Levitt and Snyder
1997, Manacorda et al. 2010, Schady 2000). The first research agenda typically argues that
macroeconomic conditions affect political preferences of some groups, mainly because these
groups evaluate different political alternatives according to certain measures like income
growth or the unemployment rate. The latter research agenda argues that voters change
their beliefs about future government behavior in response to different policies, i.e. that
different policies show the level of competitiveness of the incumbent party and the voter
interpret these as efficient (or inefficient) future behavior.
There are several benefits and differences from working with land reform in Chile that
make this paper a contribution to the literature. First, data from the Agrarian Reform
Corporation (CORA) files is available and, therefore, we know the exact amount of land
that entered into the process at each county from 1962 to 1970. The main advantage
of using this information is that there is a lot of heterogeneity among Chilean counties
level of land reform, which enable us to make good comparisons between counties affected
2
with land reform and those not affected. Second, all relevant counties are considered, and
several county characteristics can be used as covariates. Some relevant controls I use are:
income related variables (assets), supply of public goods, level of rurality, average years
of education, electoral registration, distance to trade points and region’s capital, and the
percentage of different kinds of workers (e.g. agricultural workers). Third —and this is the
main difference with several other agrarian reform process analyzed in the literature (see
Bardhan and Mookherjee 2010, for example)— the institution in charge of the agrarian
reform process (CORA) depended directly from the central, not local governments. This
puts limits to the use of land reform by local governments for political reasons, and enable us
to focus only on the central government incentives. Fourth, there was a general agreement
among political coalitions that an agrarian reform process was needed. The first political
party that developed an agrarian reform law to be presented at the Congress was the
Socialist Party (left wing), but the law enacted in 1962 was written by the Radical Party
(centre–right wing), and the process actually started under a right wing government.
My empirical strategy is to take voting data at the county level before the agrarian
reform process started (and after the introduction of the secret ballot, i.e. at the 1958
presidential elections) and use this information to control for fixed county characteristics
affecting votes for the incumbent party (e.g. ideology). Then, I estimate first-difference
OLS regressions between presidential elections in 1970 and 1958 to control for time and
county fixed effects, and control for several variables affecting government support that
vary across county and time.
Results suggest that counties with land reform increased its government support in
about 5%. This result is robust to the inclusion of a large set of relevant covariates. For
potential econometric issues I then use geographical instruments and estimate two stage
least squares. This exercise confirms first-difference OLS results. Also, different channels
and mechanisms are evaluated and I cannot reject the hypothesis that agricultural workers
were the swing voters, i.e. those who changed their voting patterns in counties with land
reform: in counties where 70% of the labor force is an agricultural worker, political support
for the government increases in 17%, while when this group is only 30% of the labor force,
government support rises in only 7%. Although it is possible that these workers evaluated
the incumbent according to land reform implementation directly, other mechanisms are also
examined. Particularly interesting is the fact that land reform is strongly correlated with an
increase in public goods provision. Agricultural workers might take this into consideration
when they decide to vote for the incumbent. However I do not rule out other potential
mechanisms.
Finally, I use the agrarian reform done by the Church during 1962 and 1963 as robust-
ness check and falsification exercise. The intention is to explore if this agrarian reform
3
increased political support for the incumbent party (the right wing, falsification exercise)
and if it increased political support for the Christian Democratic Party (political centre,
robustness check). Results do not show an increase in political support for the incumbent
party in counties where the Church distributed its land among agricultural workers. How-
ever, political support for the Christian Democratic Party did increase in these counties. I
argue these results confirm my main finding for two reasons. First, that incumbent votes
did not raise is expected because voters should not associate the agrarian reform with it.
Second, that the Christian Democratic Party obtained relatively more votes in counties
affected with land reform is consistent with the fact that the Church is closely related to
the PDC. Voters know this and might take it into consideration when they decide for which
candidate to vote for.
The rest of the paper is organized as follows. Section 2 presents the relevant historical
background in order to understand the context of this research. Section 3 presents the
theoretical mechanisms which I argue are relevant to understand the political effects of land
reform. Section 4 presents my main results under different estimation methods. Section 5
examines mechanisms and provides empirical support for the claim that agricultural workers
are the swing voters. Section 6 present a robustness check and a falsification exercise using
the Church’s agrarian reform. Finally, section 7 concludes with final remarks.
2 Chilean Rural Society and the Agrarian Reform
The influence of agriculture on Chilean society is unmeasurable, and in many ways is much
more important than mining activities such as cooper and nitrate, the other historically
important economic activities in Chile. Rural society has many special features that makes
it interesting as a subject of study in itself. As McBride (1970) puts it:
Chile’s social structure was built on land bases, and the entire life of the nation had to
be shaped in relation to land (...) The condition of each person was determined by the
ownership or not ownership of an hacienda.
This, together with Chile’s high land concentration are one of the most important charac-
teristics of rural areas. Indeed, Conning and Robinson (2007) calculate that land gini in
Chile was about 0.94 in 1965.1 Many historians hypothesized that this high land concen-
tration has its origins in colonial times (e.g. Bauer 1975 and Baraona 1960), but the lack
of data is the main reason why a more rigorous study does not exist on this subject. The
persistent high land concentration, possibly initiated at the beginning of the colonial times,
undoubtedly contributed to the formation of Chilean rural society.
1Other land gini coefficients presented in Conning and Robinson (2007) are: Argentina 0.79, Brazil 0.84,
Bolivia 0.94, Bangladesh 0.42, India 0.62, France 0.54, and United States 0.73.
4
This high land concentration was part of some kind of rural equilibrium in which rural
laborers worked for a landlord and had no opportunity to become landowners. This equi-
librium was abruptly disturb by the agrarian reform in the sixties. However, before the
sixties there was also a concern about this high concentration of property, which translated
into the creation of a government institution called Caja de Colonizacion Agrıcola in 1928
(CCA from now on, Huerta 1989, p.42-43).2 But this policy was not very effective, and
only 430 thousand physical hectares were acquired by the CCA in 30 years (1929–1958). If
we compare this number with the more than 2 millions physical hectares that entered into
the agrarian reform process between 1964–1970 it seems very small (CIDA, 1966). This
situation made it clear that a real agrarian reform could not be carried out by the CCA.
However, why was this made in the sixties and not before?
2.1 The Beginning of an Agrarian Reform
Between the creation of the CCA and the sixties, many things happened that made a more
serious agrarian reform possible. First, several political parties started to create their own
agrarian reform projects and presented them to the Congress. The first one in writing and
agrarian reform law was the socialist Marmaduque Grove in 1933, although neither this or
other projects were accepted by the Congress before the sixties (Huerta 1989, p.66). Second,
population was growing faster than agricultural production. From 1945 to 1960 the average
annual rate of growth of agricultural production was 1.8%, while the average annual rate
of population growth was about 2.2% (Tello, 1965). Chile went from being a net exporter
of agricultural products in the thirties, to have a growing trade deficit at the beginning
of the sixties. Indeed, during years 1936–1938 there was a trade surplus in agricultural
products of 1.1 millions US$, while in 1963 the annual deficit was around 124 millions US$
(Chonchol, 1976). Third, politics was ruled by a group of people with too much political
power, who also were the majority of landowners. However, this situation changed in the
fifties with the introduction of the secret ballot and the female vote. Huerta (1989) offers a
good description of this:
There is a total resistance to an structural Agrarian Reform before the fifties. The
reason is clear, it implies transmission of power, social modifications, more political
participation. Even though the agrarian problem start as an economic issue, it soon
transformed into a political problem (...) Agricultural workers have been absent as
participants of the national problems, they don’t have means of expression.
2The main objectives of this institution were to colonize State lands, make the division of this land,
intensify and industrialize agricultural production, provide credits to the beneficiaries, and afforest land
unsuitable for agricultural activities, among others.
5
Fourth, the Church’s position and the general agreement at the National Agricultural So-
ciety was that an agrarian reform was of prime necessity. Indeed, Huerta (1989) argues
that the Church’s agrarian reform before 1962 had an important effect on the national de-
bate. And fifth, the Cuban Revolution had a social impact that made redistributive policies
necessary to satisfy the social demand for it (Eckstein, 1986).
2.2 Agrarian Reform Laws
Under this scenario the agrarian reform process legally started in 1962. This process is
characterized by its two main laws that allowed the government to expropriate plots for
future redistribution.
The first law enacted was the Agrarian Reform Law #15.020 in 1962 under the right
wing government of Jorge Alessandri. This law created the Agrarian Reform Corporation
(CORA, replacing the old CCA). The CORA was a central government dependent institu-
tion in charge of the expropriation of plots. The main objectives of this law were, first, to
give access to land to those who work on it, second, to improve the living standards of the
rural population, and third, to increase agricultural production and soil productivity (Law
15.020 art. 3, Diario Oficial N.25, November 27, 1962).3
The second law (Law #16.640) was enacted in July 1967 under the centre government
of Eduardo Frei Montalva. The general agreement about the need for a more intense
land reform was reflected in the 94% of approval of this law at the Congress (Barraclough,
1971). This second law augmented the causals for expropriation of a plot and, consequently,
accelerated the agrarian reform process. Among the new causals the most important was
the one which dictated that a plot could be expropriated if it was bigger than 80 basic
irrigated hectares (BIH). This is important because a well exploited plot could also be
expropriated if it was bigger than 80 BIH after 1967. Also important was the fact that
the definition of abandonment and poor exploitation provided the CORA some discretion
for expropriating a plot. The result is that before 1967 less than 300 hundred thousand
physical hectares (PH) entered into the process, while before the 1970 presidential election
more than 2 million PH were expropriated by the CORA.
3Plots could be expropriated if: 1. the plot was abandoned and poorly exploited, 2. the CORA needed
to do irrigation works, 3. the owner of the plot had unpaid debts, 4. the owner had illegal leases, 5. the
CORA finds the plot useful, 6. the plot is mainly composed by marsh land, 7. the plot was to small and the
CORA wanted to group several small plots, 8. the plot has legally unclear ownership, 9. the plot is owned
by a corporation, and 10. if the plot is mainly composed by Araucarias (a type of tree). Basic requirements
to receive land were: 1. be Chilean, 2. be and agricultural worker, 3. be eighteen years old, 4. be skilled
in agricultural activities, 5. not to be a landowner (or own a very small plot), and 6. be married or a
householder (Law #16.640, art. 71.)
6
2.3 Politics and the Agrarian Reform under Different Governments
During the sixties there were three political coalitions: the right, the centre, and the left
wing. The right wing was composed by the Liberal and Conservative parties between 1958
and 1965, and by the National Party between 1967 and 1970. The centre was represented
by the Christian Democratic Party (PDC) and the Radical Party (PR) in 1958, but only by
the former in 1970. The left wing consisted in the union of the Socialist and the Communist
Party, and after 1969 it was also composed by the Radical Party. Therefore, when I refer
to the votes for the centre or the PDC in 1958 I implicitly mean votes either for the PDC
or the PR in 1958, but only to the votes for the PDC in 1970.
Between 1958 and 1964 the right wing government was in office with President Jorge
Alessandri. Only a few plots entered into the agrarian reform process during these years.
The only plots reformed by the CORA were the ones owned by the state (Correa et al.,
2001).4 The agrarian reform really started under the government of the Christian Democrat
Eduardo Frei Montalva, who was President of Chile between 1964 and 1970.
3 Why Land Reform Matters: Theoretical Mechanisms
This section discusses the main channels through which land reform could have affected
government support. This is important because my empirical approach in section 4 is not
able to disentangle several different mechanisms that explain my results. For a formal
discussion it is necessary to first introduce a voting scheme in which voters express their
preferences (this is motivated by the work of Fair 1978). I assume there are two different
voters (landlords and agricultural workers) and three different political candidates.
3.1 Voting Scheme
Let there be three political parties: the incumbent party from the political centre A, the
opposition party from the right wing B, and the opposition party from the left wing C. I
assume landlords do not support the left wing party and rural laborers are more likely to
vote for the left wing party (although they can also vote for the centre or right wing). I
also assume that parties A and C would like to expropriate relatively more than party B.Under this setting landlords do not have economic incentives to vote for A or C. Therefore,I will assume they always vote for the right wing candidate which, nevertheless, seems an
accurate assumption for the Chilean case.
4In fact, several historians refer to this agrarian reform period as “Reforma de Macetero” (Pot Reform),
in direct reference to the small amount of reformed land.
7
Let an agricultural worker decides for which party to vote under the following rule of
comparison among utilities:
Vote for Party k if Uk> U
m ∀k �= m, with k,m ∈ {A,B, C}
And randomizes his vote if Uk = Um. Let a worker utility be formed according to the
following process:
Ukω,c = ξω + ζc +Xc + ηcω (1)
Where ξω and ζc are agricultural worker and county fixed effects not related to land reform,
X is variable directly affected by land reform, and ηcω is a random shock with zero mean.
However, more needs to be said about what variables Xc are affected by land reform and,
at the same time, affect voting behavior. I now turn to discuss this.
3.2 Theoretical Mechanisms
If workers voted relatively more for the incumbent party in counties with land reform, why
did they do it? There are (at least) four different explanations.
1. Land reform affected some relevant variable before the election: If this happened and
workers evaluated different alternatives according to this variable they are more likely
to vote for the incumbent party in counties with land reform. This could be the case
if, for example, workers income increased relatively more in counties with land reform
(and this is caused by land reform).
2. Workers migrated to counties with land reform: If agricultural workers expect some
relevant variable to change in the future in a county with land reform, and this is
beneficial for them, they might choose to migrate to it from a county without land
reform if the benefits of doing so are bigger than the costs. This is a mechanism if
workers are more prone to vote for the incumbent (as Petras and Zeitlin 1970 suggests).
3. Workers expected some relevant variable to change in the future: This could happen
if, for example, workers assigned a higher probability to the event of becoming a
landowner under a future government of the incumbent in counties with land reform,
and they prefer being a landowner than being a landowner’s employee.
4. Workers evaluated political alternatives directly with land reform: This means that
neither present, past and/or future variables need to be affected and the incumbent
receives relatively more votes in counties with land reform. Why do workers evaluated
the incumbent according to land reform? It could be a sign of competitiveness or
signaling about concern for workers (reciprocity).
8
Although section 5 intends to show light on some of these hypothetical mechanisms, in
general it is hard to disentangle which is relatively more important because there is not
reliable data at the county level (for variables such as income) before and after land reform.
It is useful to emphasize that under this framework agricultural workers can also vote
for the left wing. In fact, they might prefer to do it if, for example, they believe their
income will be higher under a left wing government. However, I argue they do not vote in a
different way for the left wing between counties with and without land reform because they
do not associate it with the left wing. The main theoretical argument of this section is that
agricultural workers voted relatively more for the incumbent party in counties with land
reform. This could have happened if any of the above mentioned mechanisms are present.
4 Land Reform and Government Support
This section empirically explores the effects of land reform on government support. First,
I present descriptive statistics of the main variables. Then, estimates are presented un-
der three different estimation methods: differences-in-differences, first-difference OLS, and
instrumental variables.
4.1 Descriptive Statistics and Land Reform Classification
Table 1 presents summary statistics for the main variables in rural counties between regions
IV and X, the main agricultural area of Chile (see Appendix A for details). Government
support is measured as the percentage of votes the PDC obtained at the 1970 presidential
elections. The mean of this variable in 1970 is 30.7%, which is somehow smaller than the
34.5% in 1958. This reflects a shift in the electorate from the centre to the left and right
wing.
I classify 61 of the 210 (29%) counties as high expropriation counties (HEC from now
on). These counties are represented in the data with a HEC Dummy that equals 1 if more
than 7% of the county surface (in physical hectares) entered into the agrarian reform process
until August 1970 (one month before presidential election).5 Also, 149 counties (71%) are
classified as low expropriation counties (LEC from now on, i.e. the HEC Dummy equals 0).
Among these, 79 out of the 149 (53%) have at least 1 neighbor county classified as HEC.
This leaves us with 70 “isolated” counties that are not affected with land reform and do
not have a border in common with a HEC.
This table also shows that the percentage of agricultural workers increased substantially
between 1958 and 1970 (from 21% to 50%), which could be reflecting an increase in agricul-
tural activities. This increase has the same pattern in HEC and LEC, although agricultural
5Appendix B shows that results are robust to different definitions and present other robustness exercises.
9
workers were a smaller percentage of the labor force in HEC in 1958 (17% in HEC and 23%
in LEC). It is important to control for this variable because if agricultural workers have
a certain political preference and they are affected by land reform, land reform may have
had no effect on government support, and what I am capturing is the effect of a change in
labor composition. It is not important to include any other type of worker as covariate if
we believe that these are not correlated with land reform.6
Another potentially important variable which I can control for is electoral registration.
In 1958 voted 1.23 millions of voters, while in 1970 the number more than doubled to 2.92
millions (Hellinger 1978, p.255). Table 1 shows that this growth happened in both HEC and
LEC. This is important because if more people registered in HEC, and this is not caused
by land reform, I would obtain biased estimates of the effect of land reform on government
support.
Conditions and Public Goods, and Income Related variables are included as covariates
to control for two possible effects. First, to isolate the effect of land reform it is important
to control for any other government action that might be changing people’s attitude to the
government. If a county is receiving transfers from the central government between 1958 and
1970 —taxes, subsidies, public goods, or others— this could increase government support,
regardless the level of land reform in that county. Second, wage increases in one county
could be associated by its residents as good economic policy by the central government, and
might change government support.
Table 1 also shows an improvement in living standards between 1958 and 1970, measured
by increases in average education years (from 2.6 to 3.5) and literacy rate (from 67% to
73%), and increases in the percentage of houses with electricity (from 37% to 48%), hot
water (from 5% to 8%), and water supply (from 24% to 52%). It also shows an increase in
asset property measured by the percentage of houses with at least one car, television, and
radio.
4.2 Differences-in-Differences: Benchmark Estimates
Let me consider the simplest framework. If land reform was randomly assigned through
counties, we can estimate the effect of land reform on government support with differences-
in-differences with no need to control for any other variable. The identification assumption
of this method is that PDC votes are a linear function in the following way:
Vct = γc + λt + εct (2)
Vct� = γc + λt� + δ ·HECc + εct� (3)
6Indeed, results are robust to the inclusion of a wide variety of variables that reflect changes in the
percentages of different types of workers (see Table Appendix B.3, last column).
10
Where HEC is a dummy variable that equals 1 if a county c is classified as HEC and equals
0 if a county c is classified as LEC, γc is a county time-invariant fixed effect, λt is a time fixed
effect affecting all counties, and εct is a random shock with zero mean. Subscripts t and t� are
time periods before and after land reform respectively. Under these set of assumptions we
can obtain the effect of land reform on government support by taking the difference between
HEC and LEC after land reform assignment (equation 3), and subtracting the result from
the same difference before land reform assignment (equation 2). The key identification
assumption of this strategy is that the change in government support at HEC and LEC
is the same in the absence of land reform treatment, but only because some counties are
affected with land reform they differ differently after the treatment.
Estimates in Table 2 support a positive effect of land reform on government support.
The second column shows that HEC were less prone to support the PDC in 1958, but
after land reform their support for the incumbent party is the same than in LEC. If we
interpret this directly it does not exactly mean that HEC increased its government support
in absolute terms (i.e. relative to before the assignment) but rather than as a national
phenomenon counties are decreasing their political support for the PDC, but this does not
happen in HEC. To see this lets take a look at votes in non-treated counties (LEC). These
counties are voting around 6% less for the PDC, and this translates into 5% more votes for
the left wing, and 1% more votes for the right wing party.7
The main pitfall with this approach is that identification assumptions in equations (2)
and (3) could be too restrictive. There might be omitted variables correlated with land
reform and government support and, therefore, estimates in Table 2 could be biased.
4.3 Controlling for Observables: First-Difference OLS
To deal with the potential omitted variables my first strategy is to estimate first-difference
OLS regressions and to control for everything I can control for at the county level. The
obvious constraint is data availability. Thus, I take equations (2) and (3), add a matrix of
control variables at the county level Xct, and differentiate in the following way:
Vct = γc + λt + δXct + εct
Vct� = γc + λt� + δXct� + β ·HECc + εct�
Vct� − Vct = (λt� − λt) + δ(Xct� −Xct) + β ·HECc + (εct� − εct)
∆Vc = φ+ γZc + β ·HECc + ηct (4)
I take equation (4) to the data. In this case, to first-differentiate allow us to control for any
county characteristics γc that are constant over time (e.g. county ideology), the constant
7PDC support peaked around 1964, and then it decreased until 1970 (Collier and Sater 2004, p.309).
11
term φ captures the time changing preferences of the entire electorate, and Zc control for
variables that vary over county and time.8 In this case, the interpretation of the constant
term is straightforward: a negative estimate tell us that counties are voting relatively less
for the PDC, this is λt > λt� .
Table 3 present OLS estimates of equation (4). Column 1 show us the correlation
between the HEC Dummy and government support in the same way than difference-in-
difference estimates: land reform avoids a political migration of 6% from the PDC to the
left and right wing. A negative estimate of the constant term (−0.056) shows that the
electorate is migrating from the center. If we take these two estimates together we obtained
our benchmark result: political migration did not happen at HEC (0.058− 0.056 ≈ 0).
To think about counties as independent units of analysis might not be entirely appropri-
ate because counties can sometimes be very small administrative units (in terms of square
kilometers) and be close to each other. For this reason it is useful to add as a control
variable a dummy that equals one if a county is classified as LEC but has a border in
common with a HEC. The rationale behind this is that it seems naive to assume that land
reform only affects votes within the county boundaries, because sometimes these are more
de jure than de facto. Moreover, it seems intuitive to think that the effect of land reform
should be smaller or non-significant in these neighbor counties. Column 2 provides some
evidence in favor of this intuition: the effect is around half, and both effects are positive
and statistically significant.
Column 3 checks if these results are driven by differences in growth of agricultural
workers. I include agricultural workers growth because they are the biggest group in rural
counties and the most likely to migrate to a HEC. Estimates show that results are not
driven by differences in agricultural workers change. Column 4 shows that this result is
also robust to the inclusion of rurality as covariate —the change in the percentage of people
living in rural areas— and column 5 shows that this is also not driven by the fact that HEC
are voting (or enrolling at the electoral service) relatively less than LEC. To control for
potential trade or transportation policies affecting counties close to ports or trading points
I add the distance to the region’s capital and to the closest port (in hundred of kilometers).
It should not be necessary to control for these if they affect in the same way in 1958 and
1970. However, the sixties were a decade of growing commerce and decreasing transport
costs, therefore, this variable might have affected differently in 1970 and in 1958. Result is
also robust to the inclusion of these control variables.
8For example, agricultural minimum wage increased from 0.9 Escudos in 1962 to 7.5 Escudos in 1969
(Castillo and Lehmann, 1982), but this effect will be captured by the time fixed effects φ if it affects all
counties in the same way. Moreover, any change affecting all counties at different times is captured by the
constant term.
12
Columns 7 and 8 control for Conditions and Public Goods and Income Related variables.
Controlling for these variables show us that people in counties with better conditions, more
public goods, and higher income are voting relatively more for the incumbent party (more
from this in section 5). This could mean two different things. First, that land reform caused
higher income, better conditions, and more public goods in the short term, and these are
channels through which it affects government support. Second, that land reform is correlated
with these variables, and estimates in columns 1-6 are not the effect of interest, but rather
the effect of this plus the effect of omitted variables. Even if land reform did not caused
higher income, better conditions, and more public goods, it seems that in counties classified
as HEC government support increases in about 5% when we control for these variables.9
4.4 Econometric Issues: Instrumental variables
So far first-difference OLS results suggest that land reform increases government support
in about 5%. However, there might econometric problems with this estimate. In this case,
the use of an instrumental variables approach is useful for two different reasons:10
1. Controlling for Channels: If land reform causes changes in some covariates the effect
of land reform might not be 5%. If this is the case the HEC Dummy is capturing only
land reform effects not related to these covariates.
2. Measurement Error: Land reform could be measured with error for three different
reasons. First, maybe what matters is expropriation weighted by land quality, not in
physical hectares.11 Second, I take expropriation until August 1970, but I dropped a
few expropriations without date.12 Third, there could be expropriations not reported
in the CORA files. If this error is normally distributed the effect is bigger than 5%.
An instrumental variables approach solves these problems if the instruments are valid, which
depends on:
1. The presumption that the instruments are not correlated with covariates acting as
channels and the measurement error.
9This effect is bigger if more people live in rural areas (see Appendix B for details).10Although unions varied across counties and time during the sixties (Collier and Sater 2004, p.313) and
it is possible that unionized workers vote differently (see Leigh 2006 and Freeman 2003 for examples when
this happens) I assume this is not an omitted variables problem because legal procedures that allowed more
unions applied in the entire country and regions with relatively more land reform do not have more unions
(Loveman 1976, p.264).11There might be some counties classified as HEC but, if land is not very productive there, then these
should be actually classified as LEC.12Only 12 out of the 5,422 expropriations have missing date of expropriation. Among these, only 6 were
bigger than 100 physical hectares.
13
2. The need for the instrument to be strongly correlated with land reform (measured by
the HEC Dummy).
Possible covariates acting as channels are three. First, change in agricultural workers. This
is a channel if they are more likely to support the incumbent (as Petras and Zeitlin 1970
argue) and their migration is caused by land reform. Second, what I call county conditions.
This could be the case if land reform increased literacy rate or average education years.
Third, public goods, under the same above reasoning. And fourth, income related variables.
If land reform caused higher wages, and this translated into more assets, these could also
be a channel.
I use two different set of instruments: the distance from a county to the west coast
(and its square), and a dummy for landlocked counties. The first condition is met because
the instruments are not correlated with potential channels13 and because I assume the
instruments are not correlated with the measurement error.14 The rationale behind the
second condition is that the main agricultural area is geographically located in the so called
Central Valley, this is, away from the west coast and the Andes Mountains. Therefore, land
reform should be relatively more intensive in counties located in this area. In fact, the first
stage shows us that land reform was indeed more intensive here.
Table 4 present estimates using both instruments and the same result arises: there is a
positive and significant effect of land reform on the incumbent political support.15
Overall, I argue that the three different estimation methods in Tables 2, 3 and 4 provide
evidence in favor of a positive effect of land reform on government support of about 5%.
However, this effect could be bigger if some covariates are acting as channels. I now turn
to analyze this.
13Correlations between the landlocked dummy and differences in literacy rate, average education years,
houses with water supply and electricity are statistically zero (p-values are 0.14, 0.29, 0.44, 0.47 respectively).
Correlations between distance to the west coast and its square and difference in assets is also zero (p-value
0.21). Although both instruments are correlated with agricultural workers growth this is not a problem
because later on I show that this is unlikely to be a channel. However, the landlocked dummy is correlated
with assets and the distance and its square is correlated with public goods. Therefore, it is useful to use
both instruments and to compare them.14The distribution of the measurement error is unknown, therefore it is not possible to test this. However,
when I construct a dummy for counties with expropriations without date this variable is not correlated with
the instruments.15A Hausman test does not reject the null hypothesis that OLS and IV estimates are the same. Thus,
OLS estimates are more efficient.
14
5 Swing Voters and Mechanisms
This section provides a formal discussion about how voters choose among different candi-
dates —i.e. discusses mechanisms linking land reform and government support— and also
argues that agricultural workers were the swing voters —i.e. those who vote differently
in counties with and without land reform. Moreover, I discuss how agricultural workers
could have evaluated different political alternatives and what mechanisms are relatively
more important. Although it is hard to empirically answer this due to data restrictions,
some interesting correlations are provided in order to give insights about an answer.
5.1 Swing Voters
The group most positively affected by land reform could be agricultural workers. Therefore,
I suspect these could be the swing voters. Nevertheless, for a better understanding I also
analyze a large variety of different groups.
For this purpose I estimate the most complete specification and add the percentage of
different types of workers in 1970 (over labor force) and an interaction term between this
variable and the HEC Dummy. The rationale behind this strategy is to test if different
types of workers were voting relatively more for the incumbent in 1970 in counties with
land reform. The estimating equation is as follows:
∆Vc = φ + γZc + α(Wc,1970 ·HECc)
+ βHECc + ρWc,1970 + ρ1(Wc,1970 −Wc,1958) + ηct (5)
Where Wc,t stands for the percentage of a specific type of worker in county c and year t and
Zc still is the difference in covariates that vary across county and time. A positive estimate
of α means that a certain type of workers W voted relatively more for the incumbent in a
HEC.
Panel A in Table 6 present estimates of equation (5). We can see in column 1 that the
HEC Dummy is no longer statistically significant. This is in fact expected if agricultural
workers are the swing voters. Moreover, the interaction term between the HEC Dummy
and agricultural workers is statistically significant at the 5% and has the expected sign.
This estimate is interpreted in the following way: in counties classified as HEC where 70%
of the labor force is an agricultural worker, government support is 17% larger. On the
other hand, in counties classified as HEC where 30% of the labor force is an agricultural
worker, government support is only 7% larger. The rest of the columns support this finding
using several different types of workers as defined by the 1970 Housing Census. There is
no other group of workers voting relatively more for the incumbent party in counties with
land reform.
15
I also analyze if there is some political group changing its voting patterns between 1958
and 1970 among HEC and LEC. For example, if there is a political migration from the
center to the left and right wing, this migration could be different among these counties.
For this reason, I use the percentage of votes each political coalition obtained at the 1958
presidential election and estimate the following regression:
∆Vc = φ+ γZc + α1(Vkc,1958 ·HECc) + βHECc + ρ2V
kc,1958 + ηct (6)
Where Vkc,1958 is the percentage of votes party k received in 1958.
Panel B presents OLS estimates of equation (6). The first column suggests that among
the 61 counties classified as HEC government support increased relatively more where there
was lower left wing voters. The following three columns in Panel B show that there is no
other political group changing its voting patterns differently among HEC and LEC.
5.2 How Voters Evaluated Different Alternatives? Mechanisms
Following the empirical approach of Nunn (2008) and Bruhn and Gallego (2010) I now
examine different voting mechanisms. These mechanisms were already presented in section
3.2. However, mechanisms number 3 and 4 are only examined as residuals —i.e. if there is
something not explained by mechanisms 1 and 2, then these should be relevant.
According to Petras and Zeitlin (1970) agricultural workers were more prone to vote for
the PDC. Then, if they migrated relatively more to counties classified as HEC, and this
is caused by land reform, the incumbent support could have increased and, therefore, this
is a mechanism. However, column 2 in Table 5 does not show a statistically significant
correlation between land reform and the change in agricultural workers. Hence, this is
unlikely to be one of the mechanisms.
Land reform could have had an effect on some variable before the 1970 presidential
election, and through this variable could have affected voting patterns. Although many
variables could have been affected by land reform, I argue that public goods are particu-
larly important because they could be interpreted as transfers (Manacorda et al., 2010),
government spending (Levitt and Snyder 1997 and Schady 2000), or inputs for agricultural
production (as in De Gorter and Zilberman 1990). There is empirical evidence that the first
two can increase government support and an increase in productivity could have (at least
in theory) increased it too. Therefore, I focus on the correlation between the percentage of
houses with water supply and electricity with land reform. I chose these variables as proxies
for public goods because of availability from the 1960 and 1970 Housing Census. Columns
3 and 4 in Table 5 show that counties with land reform increased relatively more its elec-
tricity and water supply coverage (this could have been necessary in order to complement
land reform). This is evidence in favor of this mechanism because the correlation is strong
16
and has the expected sign.16 However, as columns 7 and 8 in Table 3 show, controlling for
changes in public goods provision still leaves an unexplained part of land reform that affects
voting behavior. Therefore, I do not rule out that changes in other variables, land reform
in itself, and changes in beliefs about what is going to happen in counties with land reform
are also mechanisms used by agricultural workers to evaluate the incumbent.
The main conclusion from this section is that the effect of land reform of government
support can be rationalized in the following way. When land reform was implemented in
a county public goods increased relatively more. Then, when agricultural workers decided
for which candidate to vote for they had a better evaluation of the incumbent (in relation
to the same worker in a county without land reform) for three different reasons. First,
they valued land reform (mechanism number 4 in section 3.2), they benefited from more
public goods (mechanism number 1), and they assigned a higher probability to the event of
becoming landowners (which is beneficial for them) or expected other variables to change
in the future (mechanism number 3).
6 Robustness Check: the Church’s Agrarian Reform
If voters change their voting patterns in response to land reform an immediate question
arises: what if land reform was not carried out by the incumbent party but rather by a
different (non-political) institution? Under my framework they should not change their
voting patterns because this does not change relative utility among voting for different
political candidates. However, this may not be entirely right if we believe that the non-
political institution is related to the right, the centre, or the left wing party.
In Chile, the Church is closely associated to the PDC (Grayson, 1969), and then, its
actions might be interpreted as information about actions of the political party. In fact,
Hudson (1994) suggests that social actions of the Church at the beginning of the sixties had
an important effect on political support for the PDC:
During the interwar years (...) the Roman Catholic Church in Chile slowly began
to espouse socially and politically more progressive positions. This more progressive
Catholicism initially had its main impact among university students, who, in the mid-
1930s under the leadership of Eduardo Frei, created a new party that in 1957 fused
with other groups to become the Christian Democratic Party. This development split
the subculture that was closer to the Catholic Church into politically conservative and
16Changes in wages are also a potential mechanism, but there is no data to be able to test this. However
column 5 in Table 5 shows that land reform is strongly correlated with the change in the percentage of houses
with radio (but not with the percentage of houses with television or cars in 1970). Changes in literacy rate
and years of education are not correlated with the HEC Dummy (not shown).
17
centrist segments. By the early 1960s, a solid majority of the church hierarchy favored
the Christian Democrats, and there was a significant shift of voter support from the
Conservative Party to the PDC. Following the new thinking in church circles, the hi-
erarchy openly embraced positions favoring land reform, much to the dismay of the
still-important minority of Catholics on the right.
The Church made its own agrarian reform during 1962 and 1963, distributing its own
plots among agricultural workers. However, we know that the Church is closely related to
the PDC. Then, not only the political support of the right wing party (the incumbent at
this time) should not have increased in counties affected by the Church’s agrarian reform,
but we also should see an increase in political support for the PDC in this counties.
In exactly the same spirit than in previous sections I take the incumbent and the PDC
political support at the 1961 Parliamentary Elections (before the Church’s agrarian reform)
and at the 1965 Parliamentary Elections (after the Church’s agrarian reform) and estimate
equation (4). The only problem in trying to recreate previous regressions is that I cannot
control for everything I would like to control for because of data availability. Furthermore,
two additional potential problems arise in this exercise. First, only a few counties were
affected with land reform (understood as more than 1% of the county’s surface begin re-
formed). Indeed, only two out of four counties where the agrarian reform was made can
be qualified as having high land reform (in the same spirit than the previous HEC). Sec-
ond, this agrarian reform was carried out only in regions VI, VII and Metropolitan (RM).
Therefore, I only take counties from these regions as the counterfactuals (or non-treated
counties).
The estates owned by the Church and assigned to rural families, with their respective
size (in physical hectares, PH) and county location, were: Alto Melipilla in Melipilla (164
PH), Los Silos de Pirque in Pirque (181 PH), Las Pataguas in Pichidegua (1,470 PH), San
Dionosio in Colbun (3,374 PH), and Alto las Cruces in Talca (340 PH). Only Pichidegua
and Colbun can be qualified as having high land reform.
Table 7 present first-difference OLS regressions to explain the incumbent political sup-
port. Different columns use different agrarian reform measures. I take the amount of land
assigned to rural families (in physical hectares) and divide it for different variables in order
to be able to compare across counties. The main difference with the most complete spec-
ification in Table 3 is that now I can only control for electoral registration, the neighbor
counties, and distances. Now, I also use dummies for regions VI and VII to control for
differences among the regions included and to control for potential selection bias. It is also
important to control for the effects of the CORA agrarian reform, in order to differentiate
the effects of the Church’s agrarian reform from the effects of the agrarian reform done by
18
the CORA.
Overall, estimates in Table 7 Panel A do not show an increase in government support in
counties with agrarian reform. In fact, they suggest that counties where the Church made
its own agrarian reform voters decrease their support for the right wing more than the
national phenomena that we actually see. In turn, this could mean that people are voting
relatively more for another party. If voters directly associated the Church and the PDC,
then an increase in PDC votes at this counties would also support my main results. Panel B
explores this possibility. Estimates show that PDC increases its support in counties where
the Church did its own agrarian reform, even when we control for electoral registration,
distances, the neighbors, and regional factors. I argue that results in both Panels are
consistent with my previous results: voters increase their support for the government in
counties with agrarian reform because they associate this with the incumbent political
actions.
7 Final Remarks
The main purpose of this paper was to study if land reform can increase the incumbent
political support. To be able to put this premise into perspective, I use a framework
that emphasizes different mechanisms linking land reform and government support. The
empirical analysis shows that using three different estimation techniques counties where
more than 7% of its surface was expropriated are more prone to vote for the incumbent
party: the incumbent obtained 5% more votes in these counties.
Also, agricultural workers seem to be the main group changing their voting patterns
between counties with and without land reform. I emphasize that several mechanisms could
be behind these results. Among these, particularly interesting is the fact that land reform
is strongly correlated with an increase in public goods provision, and is not correlated with
the change in the percentage of agricultural workers. Thus, I rule out the possibility that
a migration of agricultural workers to counties with land reform is a mechanism behind
my result. Although public goods seems to be a mechanism, there is a significant part of
the effect of land reform on government support that I cannot explain. I attribute this to
importance of land reform in itself as mechanism of evaluation —maybe because it shows
the level of competitiveness of the incumbent— and to possible changes in other relevant
variables (such as wages) before the 1970 presidential election.
Finally, I used the Church’s agrarian reform during 1962 and 1963 as robustness check
and falsification exercise that confirms my main result. The incumbent support did not
increase in counties where the Church made its own agrarian reform and votes for the party
close to the Church did increase.
19
References
Angrist, Joshua and Pischke, Jorn-Steffen: Mostly Harmless Econometrics. Prince-
ton University Press, 2009.
Baraona, Rafael: Valle de Putaendo. Santiago, 1960.
Bardhan, Pranab and Mookherjee, Dilip: �Determinants of Redistributive Politics:
An Empirical Analysis of Land Reforms in West Bengal, India�. American Economic
Review (forthcoming), 2010.
Barraclough, Solon: �Reforma Agraria: Historia y Perspectivas�. Cuadernos de
Realidad Nacional , 1971, CEREN(7).
Bauer, Arnold J.: La Sociedad Rural Chilena. Cambridge University Press, 1975.
Bruhn, Miriam and Gallego, Francisco: �Good, Bad, and Ugly Colonial Activities:
Do They Matter for Economic Development�. Economic History and Cliometrics Lab
Working Paper N.6 , 2010.
Castillo, Leonardo and Lehmann, David: �Chile’s Three Agrarian Reforms: The
Inheritors�. Bulletin of Latin American Research, 1982, 1(2), pp. 21–43.
Cerda, Rodrigo and Vergara, Rodrigo: �Business Cycle and Political Election Out-
comes: Evidence from the Chilean Democracy�. Public Choice, 2007, 132(1 and 2),
pp. 125–136.
Chonchol, Jacques: �La Reforma Agraria en Chile (1964-1973)�. Trimestre Economico,
1976, XLIII(171), p. 600.
CIDA: Chile: Tenencia de la Tierra y Desarrollo Socioeconomico del Sector Agrıcola.
Inter-American Committee for Agricultural Development (CIDA), 1966.
Collier, Simon and Sater, William: A History of Chile, 1808-2002. Cambridge Uni-
versity Press, 2004.
Conning, Jonathan and Robinson, James: �Property Rights and the Political Organi-
zation of Agriculture�. Journal of Development Economics, 2007, 82, pp. 416–447.
Correa, Sofia; Jocelyn-Holt, Alfredo; Rolle, Claudio; Vicuna, Manuel and
Figueroa, Consuelo: Historia del Siglo XX Chileno. Sudamericana, 2001.
De Gorter, Harry and Zilberman, David: �On the Political Economy of Public Goods
Inputs in Agriculture�. American Journal of Agricultural Economics, 1990, 72(1), pp.
131–137.
Eckstein, Susan: �The Impact of the Cuban Revolution: A Comparative Perspective�.
Comparative Studies in Society and History , 1986, 28(3), pp. 502–534.
Fair, Ray C.: �The Effect of Economic Events on Votes for President�. The Review of
Economic and Statistics , 1978, LX(2).
Freeman, Richard: �What Do Unions Do... to Voting?� NBER Working Paper No.
9992 , 2003.
Grayson, George W.: �Chile’s Christian Democratic Party: Power, Factions, and
Ideology�. The Review of Politics, 1969, 31(2), pp. 147–171.
Hellinger, Daniel: �Electoral Change in the Chilean Countryside: The Presidential
Elections of 1958 and 1970�. The Western Political Quarterly , 1978, 31(2), pp. 253–
273.
Hibbs, Douglas A.: Voting and the Macroeconomy. Chapter 31, pp. 565–586. Oxford
University Press, 2006.
Hudson, Rex A.: Chile: A Country Study. Washington: GPO for the Library of Congress,
1994, 1994.
Huerta, Maria Antonieta: Otro Agro para Chile: la Historia de la Reforma Agraria en
el Proceso Social y Polıtico. CISEC-CESOC, 1989.
Kramer, Gerald H.: �Short-Term Fluctuations in U.S. Voting Behavior, 1896-1964�.
The American Political Science Review , 1971, 65(1), pp. 131–143.
Leigh, Andrew: �How Do Unionists Vote? Estimating the Causal Impact of Union
Memership on Voting Behaviour from 1966 to 2004�. The Australian National University,
Centre for Economic Policy Research, Discussion Paper N.516 , 2006.
Levitt, Steven D. and Snyder, James M.: �The Impact of Federal Spending on House
Election Outcomes�. Journal of Political Economy , 1997, 105(1), pp. 30–53.
Loveman, Brian: Struggle in the Countryside: Politics and Rural Labor in Chile, 1919-
1973. Bloomington Indiana University Press, 1976.
Manacorda, Marco; Miguel, Edward and Vigorito, Andrea: �Government Trans-
fers and Political Support�, 2010.
McBride, Jorge M.: Chile: Land and Society. ICIRA, 1970.
Mikusheva, Anna and Poi, Brian: �Tests and Confidence Sets with Correct Size When
Instruments are Potentially Weak�. The Stata Journal , 2006, 6(3), pp. 335–347.
Moreira, Marcelo: �A Conditional Likelihood Ratio Test for Structural Models�.
Econometrica, 2003, 71(4), pp. 1027–1048.
Nunn, Nathan: �Slavery, Inequality, and Economic Development in the Americas: An
Examination of the Engerman-Sokoloff Hypothesis�. in E. Helpman (ed.), Institution
and Economic Performance, Harvard University Press , 2008.
Petras, James and Zeitlin, Maurice: �The Working-Class Vote in Chile: Christian
Democracy versus Marxism�. The British Journal of Sociology , 1970, 21(1), pp. 16–29.
Robinson, James and Baland, Jean-Marie: �Land and Power: Theory and Evidence
from Chile�. American Economic Review , 2008, 98(5), pp. 1737–1765.
Schady, Norbert R.: �The Political Economy of Expenditures by the Peruvian Social
Fund (FONCODES), 1991-1995�. American Political Science Review , 2000, 94(2).
Stigler, George J.: �General Economic Conditions and National Elections�. American
Economic Review , 1973, 63(2), pp. 160–167.
Tello, Carlos: �El Sector Agrıcola y el Desarrollo Economico de los Paıses
Latinoamericanos�. El Trimestre Economico, 1965.
Table
1:SummaryStatistics
BeforeLan
dReform
(1958)
After
Lan
dReform
(1970)
Sam
ple:
All
All
LEC
HEC
All
All
LEC
HEC
Mean
St.
Dev.
Mean
Mean
Difference
Mean
St.
Dev.
Mean
Mean
Difference
MainVariables
PDC
votes
0.345
(0.116)
0.361
0.303
0.058***
0.307
(0.065)
0.307
0.306
0.001
HEC
Dummy
——
——
—0.289
(0.454)
——
—
HighExp
ropriationNeigh
bor
——
——
—0.379
(0.486)
0.533
——
AgriculturalWorkers
0.211
(0.139)
0.229
0.167
0.062***
0.507
(0.159)
0.508
0.505
0.003
Rurality
0.695
(0.179)
0.685
0.720
-0.035
0.600
(0.188)
0.593
0.616
-0.023
Electoral
Registration
2.564
(2.541)
2.570
2.551
0.019
5.183
(6.054)
5.292
4.916
0.376
Distance
toRegion’s
Cap
ital
0.683
(0.396)
0.677
0.696
-0.019
0.683
(0.396)
0.677
0.696
-0.019
Distance
toClosest
Port
1.041
(0.601)
1.089
0.925
0.163*
1.041
(0.601)
1.089
0.925
0.163*
Con
ditionsan
dPublicGoo
ds
Education
2.653
(0.652)
2.654
2.651
0.003
3.502
(0.648)
3.507
3.490
0.017
Electricity
0.373
(0.186)
0.359
0.408
-0.489*
0.482
(0.188)
0.462
0.531
-0.068**
Hot
Water
0.049
(0.043)
0.051
0.043
0.008
0.084
(0.065)
0.085
0.079
0.006
Literacy
0.672
(0.066)
0.673
0.668
0.005
0.734
(0.052)
0.735
0.733
0.002
Water
Supply
0.244
(0.157)
0.247
0.238
0.008
0.521
(0.155)
0.515
0.537
-0.022
IncomeRelated
Cars
——
——
—0.055
(0.024)
0.052
0.061
-0.009
Television
——
——
—0.046
(0.054)
0.042
0.054
-0.011
Rad
io0.296
(0.158)
0.285
0.323
-0.039
0.638
(0.119)
0.620
0.683
-0.064***
Notes:
Significancelevelforcolumn
labeled
“Difference”:***
p<0.01,**
p<0.05,*
p<0.1.Summary
Statisticsfor210
non-urban
countiesbetween
regionsIV
and
X(All).HEC:High
expropriation
counties,wheremorethan
7%
ofthecounty
surfaceentered
into
theagrarian
reform
processbeforeAugust1970.LEC:
Low
expropriation
counties,wherelessthan
7%
ofthecounty
surfaceentered
into
theagrarian
reform
processbeforeAugust1970.SeeAppendix
Aforsources
and
definition
ofvariables.
Table
2:Differen
ces-in-D
ifferen
cesEstim
ates
Presidential
Election19
58Presidential
Election19
70
Con
trol
Treated
Diff
Con
trol
Treated
Diff
Diff-in-D
iff
(%vo
tes)
(%votes)
(λt 1−
λt 0)
(%votes)
(%votes)
(λt 1−λt 0+δ)
δ
LeftW
ing
28.9
31.5
2.7
33.6
32.9
-0.7
-3.4
(0.99)
(1.54)
(1.83)
(0.99)
(1.54)
(1.83)
(2.58)
Center
36.3
30.3
-5.9
30.7
30.6
-0.1
5.8*
**
(0.76)
(1.18)
(1.40)
(0.76)
(1.18)
(1.40)
(1.98)
Right
Wing
34.9
38.2
3.3
35.7
36.5
0.8
-2.5
(0.89)
(1.39)
(1.65)
(0.89)
(1.39)
(1.65)
(2.33)
Left+
Center
65.1
61.9
-3.3
64.3
63.5
-0.8
2.5
(0.89)
(1.39)
(1.65)
(0.89)
(1.39)
(1.65)
(2.3)
Notes:
Significance
level:
***p<0.01
,**
p<0.05
,*p<0.1.
210non
-urban
counties
between
region
sIV
and
X(61treated
counties
and
149control
counties).
See
Appen
dix
Aforsources
and
defi
nition
ofvariab
les.
Greek
letterscomefrom
thefollow
ingequations:
Vct
=
γ c+
λt+
δXct+
ε ctan
dVct�
=γ c
+λt�+
δXct�+
ε ct�,wheret=
1958
andt�=
1970
.
Table 3: OLS Results — Robustness to Control Variables
Dependent variable: PDC votes in 1970 minus PDC votes in 1958
(1) (2) (3) (4) (5) (6) (7) (8)
HEC Dummy 0.058*** 0.084*** 0.060*** 0.054** 0.054** 0.067*** 0.069*** 0.051**
(0.019) (0.021) (0.022) (0.022) (0.022) (0.021) (0.021) (0.021)
High Expropriation Neighbor 0.048** 0.029 0.027 0.026 0.032* 0.032 0.031
(0.020) (0.021) (0.021) (0.021) (0.020) (0.020) (0.020)
Agricultural Workers 0.224*** 0.241*** 0.242*** 0.211*** 0.219*** 0.198***
(0.054) (0.054) (0.054) (0.054) (0.063) (0.061)
Rurality -0.291** -0.289** -0.486*** -0.511*** -0.435**
(0.143) (0.143) (0.149) (0.162) (0.168)
Electoral Inscription 0.002 0.003 0.003 -0.008
(0.006) (0.008) (0.007) (0.007)
Distance to Regions’ Capital -0.064*** -0.064*** -0.042*
(0.022) (0.023) (0.022)
Distance to closest Port 0.054*** 0.057*** 0.064***
(0.013) (0.015) (0.013)
Constant -0.056*** -0.081*** -0.133*** -0.163*** -0.163*** -0.191*** -0.150*** -0.341***
(0.010) (0.013) (0.016) (0.022) (0.022) (0.034) (0.048) (0.074)
Conditions and Public Goods No No No No No No Yes Yes
Income Related No No No No No No No Yes
Counties 210 210 210 210 210 210 210 210
R2 0.044 0.071 0.141 0.157 0.157 0.240 0.253 0.343
Notes: Robust standard errors in parenthesis. Significance level: *** p<0.01, ** p<0.05, * p<0.1. HEC
Dummy equals 1 if more than 7% of the county surface entered into the agrarian reform process before
August 1970 and 0 otherwise. See Appendix A for sources and definition of variables.
Table 4: Instrumental Variables
Dependent variable: PDC votes in 1970 minus PDC votes in 1958
(1) (2) (3)
HEC Dummy 0.156* 0.140* 0.145*
(0.086) (0.083) (0.076)
High Expropriation Neighbor 0.082* 0.074* 0.077*
(0.044) (0.042) (0.039)
Electoral Inscription -0.008 -0.008 -0.008
(0.007) (0.006) (0.006)
Agricultural Workers 0.131* 0.141* 0.138*
(0.078) (0.082) (0.077)
Rurality -0.422** -0.424** -0.423**
(0.173) (0.171) (0.171)
Distance to Regions’ Capital -0.054** -0.053** -0.053**
(0.027) (0.026) (0.026)
Distance to closest Port 0.071*** 0.070*** 0.070***
(0.014) (0.014) (0.014)
Constant -0.303*** -0.309*** -0.307***
(0.084) (0.081) (0.081)
Conditions and Public Goods Yes Yes Yes
Income Related Yes Yes Yes
Counties 210 210 210
F-test excluded instruments 9.935 6.092 4.570
Over-identification test (p-value) – 0.751 0.944
CLR (p-value for HEC Dummy) 0.097 0.130 0.086
Hausman test (p-value) 0.208 0.268 0.198
Notes: Robust standard errors in parenthesis. Significance level: *** p<0.01, ** p<0.05,
* p<0.1. The HEC Dummy equals 1 if more than 7% of the county surface entered into
the agrarian reform process before August 1970 and equals 0 if it does not. Instrument
in the first column is the distance from a county and its square, in column 2 a dummy
for landlocked counties, and in column 3 both. See Appendix A for sources and definition
of variables. Sargan over-identification tests are calculated from regressions with standard
errors and Conditional Likelihood Ratio (CLR) from Moreira (2003) using Stata module
from Mikusheva and Poi (2006) to show that weak instruments is not a problem.
Table 5: Possible Mechanisms linking Land Reform and Government Support
Dependent variable: Difference between 1970 and 1960 in
% Agricultural Workers % Houses with
Water Electricity Radio
Supply
HEC Dummy 0.037* -0.011 0.036*** 0.017** 0.028***
(0.021) (0.013) (0.009) (0.007) (0.010)
Distance to Regions’ Capital -0.009 -0.042** -0.012 -0.007 0.013
(0.026) (0.016) (0.011) (0.010) (0.012)
Distance to closest Port 0.061*** 0.032*** -0.019** -0.022*** -0.009
(0.017) (0.008) (0.008) (0.007) (0.009)
Rurality in 1960 0.711*** -0.352*** 0.038 0.185***
(0.033) (0.074) (0.041) (0.038)
Agricultural workers in 1960 -0.539*** -0.826***
(0.069) (0.034)
Houses with water supply in 1960 -0.513***
(0.079)
Houses with electricity in 1960 -0.039
(0.035)
Houses with radio in 1960 -0.258***
(0.053)
Constant 0.341*** -0.026 0.664*** 0.120*** 0.282***
(0.031) (0.022) (0.071) (0.041) (0.046)
Observations 210 210 210 210 210
R-squared 0.237 0.782 0.239 0.114 0.537
Notes: Robust standard errors in parenthesis. Significance level: *** p<0.01, ** p<0.05, * p<0.1. The
HEC Dummy equals 1 if more than 7% of the county surface entered into the agrarian reform process
before August 1970 and equals 0 if it does not. See Appendix A for sources and definition of variables.
Table
6:SwingVoters
Dep
endentvariab
le:PDC
votesin
1970
minusPDC
votesin
1958
Pan
elA
:DifferentTyp
esof
Workers
Agricultural
Clerks
Craftsan
dTrades
Plant
andMachine
Professionalsan
dService
andSalesman
Workers
over
Workers
over
Workers
over
Workers
over
TechnicianWorkers
Workers
over
Lab
orForce
Lab
orForce
Lab
orForce
Lab
orForce
over
Lab
orForce
Lab
orForce
HEC
Dummy
-0.068
0.078**
0.126***
0.060
0.096**
0.115***
(0.054)
(0.037)
(0.046)
(0.053)
(0.038)
(0.043)
HEC
Dummy×
Typ
eof
Workers
0.236**
-1.566
-0.577**
-0.329
-1.429*
-0.936*
(0.105)
(1.277)
(0.287)
(1.044)
(0.837)
(0.517)
R2
0.359
0.368
0.359
0.362
0.391
0.366
Pan
elB
:DifferentPolitical
Groups
LeftW
ing
PDC
Right
Wing
Rad
ical
Party
HEC
Dummy
0.152***
-0.017
0.001
0.035
(0.057)
(0.034)
(0.031)
(0.031)
HEC
Dummy×
Political
Group
-0.347**
0.037
0.092
-0.169
(0.160)
(0.095)
(0.080)
(0.152)
R2
0.371
0.795
0.624
0.647
Con
ditionsan
dPublicGoo
ds
Yes
Yes
Yes
Yes
Yes
Yes
IncomeRelated
Yes
Yes
Yes
Yes
Yes
Yes
Other
Con
trols
Yes
Yes
Yes
Yes
Yes
Yes
Cou
nties
210
210
210
210
210
210
Notes:Rob
ust
stan
darderrors
inparen
thesis.Significance
level:
***p<0.01
,**
p<0.05
,*p<0.1.
TheHEC
Dummyequals1ifmorethan
7%of
thecounty
surfaceen
teredinto
theag
rarian
reform
process
beforeAugu
st19
70an
dequals0ifit
does
not.See
Appen
dix
Aforsources
anddefi
nitionof
variab
les.
Table 7: The Church’s Agrarian Reform
Expropriation Expropriation Expropriation HEC Expropriation
over County over Agricultural over total Dummy over total
Surface Surface Workers Votes
Panel A : Dependent variable: Right wing votes in 1965 minus Right wing votes in 1961
Church Agrarian Reform -0.185 -0.157 -0.129*** -0.103*** -0.093***
(0.117) (0.097) (0.023) (0.031) (0.017)
Church Agrarian Reform Neighbor 0.078* 0.077 0.075 0.073 0.075
(0.046) (0.046) (0.047) (0.047) (0.047)
Agrarian Reform 1965 0.118 0.131 0.090 0.303 0.092
(0.274) (0.275) (0.258) (0.366) (0.258)
R2 0.312 0.318 0.334 0.341 0.334
Panel B : Dependent variable: PDC votes in 1965 minus PDC votes in 1961
Church Agrarian Reform 0.184* 0.153* 0.127*** 0.089*** 0.093***
(0.103) (0.086) (0.024) (0.022) (0.017)
Church Agrarian Reform Neighbor -0.019 -0.018 -0.016 -0.015 -0.016
(0.029) (0.029) (0.030) (0.030) (0.030)
Agrarian Reform 1965 0.620*** 0.610*** 0.649*** 0.469* 0.647***
(0.194) (0.193) (0.197) (0.243) (0.197)
R2 0.254 0.259 0.276 0.274 0.275
Distances Yes Yes Yes Yes Yes
Region Dummies Yes Yes Yes Yes Yes
Counties 74 74 74 74 74
Notes: Robust standard errors in parenthesis. Significance level: *** p<0.01, ** p<0.05, * p<0.1. Esti-
mates for 73 non-urban counties in regions VI, VII and Metropolitan (RM). Dependent variable in Panel
A (B) is the percentage of votes the Conservative and Liberal parties (Christian Democratic Party) ob-
tained at the 1965 Parliamentary Elections minus the same percentage in the 1961 Parliamentary Elections.
Church Agrarian Reform is the amount of physical hectares distributed to agricultural workers over county
surface (in physical hectares) during 1962 and 1963. See Appendix A for sources and definition of variables.
A Data Construction
This appendix shows data construction from the CORA files, definitions and sources for
the main variables, and argues why only 210 counties are considered.
A.1 Agrarian Reform Index
There is information about the amount of expropriated land over surface in the county,
where both measures are in physical hectares (PH) for the 257 counties in the agrarian
reform database. Therefore, the de facto agrarian reform intensity index at county c (ARIc)
I consider in the empirical section has the following mathematical form:
ARIc,t =
�p∈c (Expropriated PH of plot p−Non Agrarian Transferences from plot p)t
(PH Surface of County c)t
Where p ∈ c recognized that there are many plots in a single county, and the numerator
captures the actual amount of land reform net of redistributed land with non agrarian
objectives.17 Because the agrarian reform process started in 1962 and finished in 1980 I
constructed an index until August of 1970, 1 month before the presidential election.
A.2 Counties between Regions IV and X
Land reform is intended to affect rural counties where agriculture is an important economic
activity. Therefore, my focus is only on 210 non-urban counties between regions IV and X,
the main agricultural area of Chile (see Figure B.2). As supporting evidence for this decision
lets consider arable hectares (suitable land for growing crops) across Chilean regions: in 1955
there were 5.5 million arable hectares between regions IV and X, and only 294 thousand
arables hectares in regions I, II, III, XI, and XII. (CIDA 1966, p.24). Thus, focus on rural
counties in the aforementioned regions seems natural to analyze the effects of land reform
on government support.
Excluded urban counties between regions IV and X are: La Serena, Vina del Mar, Quinta
Normal, Santiago, Maipu, San Miguel, Quilicura, Renca, Barrancas, Maestranza Conchalı,
Providencia, Nunoa, La Reina, La Cisterna, Puente Alto, Las Condes, La Florida, La
Granja, Rancagua, Lota, Talca, Concepcion, Penco, Coronel, and Temuco.
17Non-agrarian objectives are land transferences to non-agrarian state companies, sport clubs, munici-
palities, education ministry and other ministries. 6.6% of the expropriated land at the national level had
non-agrarian objectives.
Table Appendix A.1: Definition of Variables and Sources
Variable Definition and Source
Dummy High Expropriation Dummy equals 1 if more than 7% of the county surface was expropriated
before August 1970 (Agrarian Reform Corporation files).
High Expropriation Neighbor Identification of borders in common across counties
with Cartographica (GIS) using data from GIS Chile
(http://www.rulamahue.cl/mapoteca/catalogos/chile.html).
Agricultural Workers Percentage of “Skilled Agricultural” workers over labor force (1970 and
1960 Housing Census, IPUMS).
Rurality Percentage of people living in rural areas (1970 and 1960 Housing Cen-
sus, IPUMS).
Electoral Registration Number of voters in 1970 minus the number of voters in 1958 over voters
in 1958, Electoral Service (SERVEL)
PDC votes Percentage of votes for the Christian Democratic Party and the Radical
Party in 1958 and percentage of votes for the Christian Democratic Party
in 1970 (Electoral Service, SERVEL).
Right wing votes Percentage of votes for Jorge Alessandri in 1958 and 1970 (Electoral
Service, SERVEL).
Left wing votes Percentage of votes for Salvador Allende and Antonio Zamorano in 1958
and percentage of votes for Salvador Allende in 1970 (Electoral Service,
SERVEL).
Distance to Region’s Capital From a county’s centroid to the capital’s centroid using Google Maps
for latitude and longitude locations and Stata’s vincenty command for
calculations.
Distance to closest Port From a county’s centroid to the capital’s centroid using Google Maps
for latitude and longitude locations and Stata’s vincenty command for
calculations.
Dummy for Landlocked Dummy equals 1 if the county is landlocked. Iden-
tification using Cartographica with GIS Chile data
(http://www.rulamahue.cl/mapoteca/catalogos/chile.html)
Conditions and Public Goods Average years of education, percentage of people who know how to read
and write, and percentage of houses with electricity, water supply, and
hot water (1970 and 1960 Housing Census, IPUMS).
Income Related Percentage of houses with at least 1 car and 1 television (1970 Housing
Census, IPUMS) and with at least 1 radio (1960 and 1970 Housing
Census, IPUMS).
Church Agrarian Reform Counties where the Church distributed its own plots among agricultural
workers (Huerta, 1989).
Church Agrarian Reform Neighbor Identification of borders in common across counties with
Cartographica (GIS) using GIS data from GIS Chile
(http://www.rulamahue.cl/mapoteca/catalogos/chile.html).
B Robustness Exercises
B.1 Land Reform Measures
This subsection presents estimates of the same regressions in Table 3 but using different
land reform measures.
Table Appendix B.2 shows that results are robust to different measures of expropriation
by estimating the most complete OLS specification (column 8) with different expropria-
tion variables. All expropriation measures are represented by the total amount of physical
hectares that entered into the agrarian reform process before August 1970 (one month be-
fore the elections). For example, the second column takes as denominator the exploitations
surface (in physical hectares), i.e. the amount of land used in economic activities (agri-
culture, stockbreeding), while the third column takes as denominator the total amount of
physical hectares in the county, regardless if it was use for economic activities. I use a HEC
dummy because its interpretation is easier and all the other measures are more likely to be
measured with error. If anything, I take the most conservative variable (measured by its
statistical significance).
B.2 Using Different Sub-samples
Table Appendix B.3 present two different exercises. First, the first eight columns show that
my main result is not driven by any particular region. Each column represents a different
OLS regression of equation (4) using different restricted samples. Second, the last column
control for changes in the percentage of different types of workers over the labor force (see
Table 6 for more details). Results are also robust to the inclusion of these covariates.
Finally, Table Appendix B.4 includes the percentage of the county surface expropriated
under each of the four most used expropriation causals.18 Results in this table show that
counties where most of the plots were expropriated under causals number 3 and 6 seem to be
changing its voting patterns relatively more. This is in fact intuitive because expropriation
causal number 4 is not widely used as the other three —and thus it is difficult to cause a
big effect on voting patterns— and expropriation causal number 10 is related to a plot that
is offered by the owner.
18Indeed, 32% of the 2 millions of physical hectares expropriated between 1967 and 1970 were expropriated
using expropriation causal number 3 (Plot is bigger than 80 basic irrigated hectares), 1% using expropriation
causal number 4 (Plot is inefficient or abandoned), 29% using expropriation causal number 6 (Plot is owned
by a corporation), and 38% using expropriation causal number 10 (Plot is offered by the owner to the
CORA).
B.3 Interactions and Econometric Exercises
For a better understanding of results I also explore some interactions and perform some
econometric exercises. Column 1 in Table Appendix B.5 uses a HEC Dummy that equals
1 if the county is affected with land reform before 1965 as a proxy for the original HEC
Dummy. As I already mentioned, land reform before 1965 may not matter for several
reasons. First, as section 2 argues, the main expropriation causals used before 1965 were
completely different from those used after 1967. And second, we are still far away from
upcoming presidential elections. Estimates in column 1 shows that this variable does not
affect government support, a result in line with the one presented in columns 7 and 8 in
Table 3. Thus, it is possible that land reform had different effects in a dynamic setting,
where the effect is bigger the closer we are from upcoming elections.
I also explore if there was an heterogeneous effect in counties with different rurality
levels, understood as the percentage of people living in rural areas. Even though I am
working with non-urban counties, rurality level varies within these. It seems intuitive to
think that the effect should be bigger in counties with more rural population, because the
percentage of the electorate affected by land reform is bigger. Column 2 in Table Appendix
B.5 explores this possibility and suggest that the effect of land reform indeed varies with
the level of rurality. For example, in a HEC where 50% of the population live in rural areas,
government support increases in 9% (relative to LEC). On the other hand, in a HEC where
rural population is 90%, government support rises in 17%. Both interpretations consider
that land reform does not have an independent effect on the dependent variable, as its
statistical significance suggests.
As I already mentioned when I justified the inclusion of the dummy for a LEC that is
neighbor of a HEC as covariate, counties are small units of analysis, and land reform in one
county could have affected government support in a neighbor county. To further explore this
effect I estimate the most complete OLS specification, but using as dependent variable the
difference in PDC votes in the closest county. Distance to the closest county is measured in
kilometers from a county’s centroid to the neighbor minus the average distance between two
neighbor counties —the average distance between two counties is 17 kilometers. Column 3
shows that the effect for the average neighbor is an increase in government support of 6%
and that this effect is smaller the farther the neighbor county is and bigger the closest it is.
If a neighbor county’s distance to a HEC is 15 kilometers more than the average (the actual
case of the farthest county), the estimates suggest the effect of land reform on government
support is zero in the neighbor county. On the other hand, if a LEC is very close to a HEC
—say, 10 kilometers less than the average— government support seems to increase in about
10%. Column 4 includes the other two relevant distances as covariates and the significance
of the interaction is now significantly different from zero only at the 14%. Now, if land
reform is the only variable that has a spatial relevance —i.e. affects other counties besides
its own— then, under this setting the omitted variables problem is no longer relevant. The
rationale of this assertion relies on the fact that if this is true, then these omitted variables
are correlated with the HEC Dummy in its own county, not the neighbor’s. However, to
test if this is indeed the case we would need the omitted variables which, of course, are not
available.
Table Appendix B.2: Robustness to Different Expropriation Measures
Dependent variable: PDC votes in 1970 minus PDC votes in 1958
HEC Expropriation Expropriation
Dummy over Exploitations over County
Surface Surface
Expropriation 0.051** 0.116*** 0.112***
(0.021) (0.038) (0.038)
Conditions and Public Goods Yes Yes Yes
Income Related Yes Yes Yes
Other controls Yes Yes Yes
Counties 210 210 210
R2 0.343 0.355 0.352
Notes: Robust standard errors in parenthesis. Significance level: *** p<0.01, ** p<0.05, * p<0.1.
Full specification (Table 3, column 8).
Table
Appendix
B.3:Rob
ustnessexercise
excludingcountiesfrom
specificregion
s
Dep
endentvariab
le:PDC
votesin
1970
minusPDC
votesin
1958
Excluded
Region:
IVV
VI
VII
VIII
IXX
R.M
.Non
e
HEC
Dummy
0.053**
0.062***
0.047**
0.036*
0.044*
0.050**
0.044*
0.042**
0.046**
(0.023)
(0.022)
(0.024)
(0.022)
(0.024)
(0.023)
(0.023)
(0.021)
(0.022)
HighExp
ropriationNeigh
bor
0.026
0.037*
0.031
0.022
0.035*
0.032
0.035
0.028
0.028
(0.021)
(0.022)
(0.021)
(0.021)
(0.021)
(0.023)
(0.022)
(0.020)
(0.021)
AgriculturalWorkers
0.195***
0.158**
0.149**
0.173**
0.294***
0.246***
0.162**
0.208***
0.202***
(0.067)
(0.069)
(0.071)
(0.069)
(0.068)
(0.068)
(0.074)
(0.062)
(0.067)
Con
ditionsan
dPublicGoo
ds
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
IncomeRelated
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Distances
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
DifferentWorkers
No
No
No
No
No
No
No
No
Yes
Observations
197
182
181
184
173
188
179
193
210
R2
0.353
0.395
0.310
0.316
0.330
0.357
0.377
0.388
0.352
Notes:Rob
ust
stan
darderrors
inparenthesis.Significance
level:
***p<0.01
,**
p<0.05
,*p<0.1.
Table Appendix B.4: Expropriation under different Causals
Dependent variable: PDC votes in 1970 minus PDC votes in 1958
(1) (2) (3) (4) (5)
HEC Dummy 0.026 0.048** 0.043** 0.051** 0.030
(0.022) (0.022) (0.021) (0.022) (0.023)
Expropriation under Causal N.3 0.307*** 0.289***
(0.074) (0.078)
Expropriation under Causal N.4 0.106 -0.082
(0.219) (0.195)
Expropriation under Causal N.6 0.338*** 0.164*
(0.120) (0.091)
Expropriation under Causal N.10 -0.001 -0.055
(0.059) (0.061)
Conditions and Public Goods Yes Yes Yes Yes Yes
Income Related Yes Yes Yes Yes Yes
Other controls Yes Yes Yes Yes Yes
Observations 210 210 210 210 210
R2 0.378 0.344 0.359 0.343 0.383
Notes: Robust standard errors in parenthesis. Significance level: *** p<0.01, ** p<0.05, * p<0.1.
Table Appendix B.5: Interactions and Falsification Exercises
Dependent variable is PDC votes in 1970 minus PDC votes in 1958 from:
Own county Closest County
(1) (2) (3) (4)
HEC Dummy 0.001 -0.064 0.060*** 0.074***
(0.025) (0.051) (0.022) (0.021)
HEC Dummy × Rurality in 1970 0.188**
(0.079)
HEC Dummy × Distance to closest County -0.004** -0.003
(0.002) (0.002)
Distance to Regions’ Capital -0.036 -0.037 -0.021
(0.023) (0.023) (0.021)
Distance to closest Port 0.060*** 0.064*** 0.086***
(0.014) (0.013) (0.015)
Controls Yes Yes Yes Yes
Conditions and Public Goods Yes Yes Yes Yes
Income Related Yes Yes Yes Yes
Counties 210 210 210 210
R2 0.323 0.354 0.247 0.350
Notes: Robust standard errors in parenthesis. Significance level: *** p<0.01, ** p<0.05, * p<0.1. Ex-
propriation causals are: Plots larger than 80 BIH (causal N.3), Plots are inefficient or abandoned (Causal
N.4), Owners of the plot are juridical people (causal N.6), Plots were offered to the CORA by the owner
(causal N.10).