POLITICAL ECONOMY OF LAND REFORMS IN WEST BENGAL

1978-981

Pranab Bardhan2 and Dilip Mookherjee3

This version: May 4, 2005

Abstract

We examine determinants of political will of local governments to implement land

reforms in a longitudinal sample of 88 villages in the Indian state of West Bengal.

The evidence shows a inverted-U pattern between land reform and control of local

governments by a Left party coalition, inconsistent with both polar hypotheses that

political will is determined by ideology or electoral opportunism alone. The empiri-

cal patterns are consistent with a hybrid competition-cum-ideology-cum-moral hazard

model of greater incentive for competing party officials to implement land reforms when

elections are more contested.

1We thank the MacArthur Foundation Inequality Network for funding the data collection. Sankar Bhau-

mik and Sukanta Bhattacharya of the Department of Economics, Calcutta University led the village survey

teams that collected the data. Indrajit Mallick helped us obtain the election data. We are grateful to

Debu Bandyopadhyay, Abhijit Banerjee, Partha Chatterjee, Esther Duflo, Andy Foster, Michael Kremer,

Kevin Lang and Kaivan Munshi for useful discussions. Alfredo Cuecuecha, Nobuo Yoshida, Amaresh Ti-

wari, Satadru Bhattacharya and especially Monica Parra Torrado provided excellent research assistance.

Mookherjee thanks the John Henry Simon Guggenheim Foundation for funding a sabbatical year when part

of this research was conducted. The paper has benefited from the comments of seminar participants at Ja-

davpur, MIT, PennState, Stanford, Toulouse, World Bank, MacArthur Inequality network, and the Center

for Studies in Social Science, Calcutta.2Department of Economics, University of California, Berkeley3Department of Economics, Boston University

1

1 Introduction

In many developing countries, land reforms have significant potential for simultaneously

reducing poverty and promoting agricultural growth. For instance, there is evidence that

small farms are more productive than large farms (e.g., Bardhan (1973), Berry and Cline

(1979), Binswanger and Rosenzweig (1986, 1993), Binswanger, Deininger and Feder (1993)),

that owner-cultivated farms are more productive than tenant farms (Bell (1977), Sen (1981),

Shaban (1987)), both of which imply agricultural output would rise following redistribution

of land. Banerjee, Gertler and Ghatak (2002) argue that protection of sharecroppers against

eviction and regulating sharecropping contracts in the Indian state of West Bengal in the late

1970s caused significant growth in agricultural yields. Besley and Burgess (2000) find that

implementation of land reforms (particularly with respect to tenancy protection legislation)

in Indian states between 1958 and 1992 led to significant reductions in rates of rural poverty.

Yet the extent of land reforms enacted typically remains small relative to what could

potentially be achieved. The causes are rooted mainly in the nature of governance: lack of

political will, the power of landed interests, and formidable legal and administrative barriers.

The latter stem from poor state of land records, pervasiveness of legal loopholes and legal

systems ill-equipped to deal with large volumes of litigation. It can be argued, however,

that persistence of legal and administrative barriers owe ultimately to lack of political will:

when governments really do intend to carry out land reforms they can improve the land

records, push through legislative reforms to close loopholes, and pursue necessary litigation.

From this standpoint political will is the fundamental sine qua non. If so, it is important

to understand the institutional determinants of political will of governments to carry out

land reform.

Post-Independence India represents a interesting instance of stated objectives of land

reform as a central priority of elected governments since the 1950s, where the actual progress

has been minor by comparison. The key problems have been in implementation of the stated

objectives. Agriculture being a state subject in the Indian Constitution, implementation of

land reforms has been left to the state governments, where landed interests often play an

important role. It is generally admitted that the pace of implementation has been largely a

2

matter of political will of the corresponding state governments (see, for example, the review

of the Indian land reform experience by Appu (1996)).

This paper focuses on the experience of West Bengal, an Indian state with a democrat-

ically elected Left Front government continuously since 1977, which is widely believed to

have enacted a successful land reform program during this period. From 1978 onwards,

they created a system of mandatory election to local governments with a five year term,

and devolved to them significant responsibility for implementation of the reforms. The

principal reform initiatives comprised land redistribution (patta program) and registration

of sharecroppers (barga program). Our analysis is based on a longitudinal dataset we have

assembled of 88 villages spread throughout the state spanning three decades.

West Bengal politics has been characterized by competition between two leading parties,

the Left Front coalition, and the Congress party. The Congress party dominated state

politics until the mid-1960s, whose traditional base in rural areas has been dominated by

big landlords. Since the late 1970s, the Left Front has dominated both the state legislature

and local governments. It represents a constituency dominated by landless and marginal

landholders, with big landowners stated as the ‘class enemy’ in the party’s ideology. Given

that the land reform program of West Bengal appeared to take off only after the advent

of the Left Front government in the late 1970s, a natural hypothesis is that differences

in redistributive ideology of elected governments accounted for the post-70s acceleration

in reforms. An alternative and more cynical view is that the compulsions of electoral

competition combined with opportunistic power-seeking motives were the real driving forces.

Given large degrees of landlessness, high population density and ‘land hunger’ emanating

from lack of adequate employment opportunities in industry, land reforms represented a

way for the Left parties to build up a large electoral constituency among the poor.

These alternate views correspond to different political economy theories. The ideology

based hypothesis (dating back to Lipset (1960) and Wittman (1973)) states that parties

or politicians have intrinsic policy preferences derived from their ideology (defined broadly

to include interests of constituents they represent). Combined with the assumption that

candidates cannot commit to their policy platforms, and that they ignore implications of

3

current policy choices for future re-election prospects, such a theory implies that policy

choices of elected candidates are entirely ‘ideology’ determined. Accordingly predicting

policy choices translates into predicting electoral success of parties with different ideologies.

In contrast, the theory of Downs (1957) stresses the role of competition and electoral

opportunism, by assuming that candidates have no intrinsic policy preferences. Policies are

chosen in order to secure votes in current or future elections. In a two candidate setting it

predicts policy convergence: both candidates select the same policy owing to their common

vote-maximization objective. Such a theory thus predicts policy choices are independent

of the identity of the winning candidate, and are explained instead by preferences and

turnout patterns of voters. In contrast to the ideology-based theory which stresses the role

of political parties, the Downsian view stresses the role of competitive electoral incentives:

that political will is driven ultimately by policy preferences of voters, not parties.

Simple correlations with the share of seats in local governments (gram panchayats (GP))

captured by the Left Front — either across villages over the entire twenty year period,

or within villages across successive elected local governments — show no support for the

ideology hypothesis. While the correlations are generally statistically insignificant, the data

shows an inverted U-pattern in both cross-section and longitudinal data, with the downward

sloping portion prevailing over most of the sample. The absence of a significant relationship

between Left share of local government seats and land reforms implemented are consistent

with the Downsian model.

However this inference is subject to two significant caveats. First, there is potential

endogeneity bias: the success of the Left is presumably jointly determined along with policies

implemented. Second, the inverted-U relationship if at all empirically significant would need

to be explained, as it is not consistent with the predictions of a standard Downsian model.

To address the first problem, we use instruments for observed success of the Left Front

in local elections. It turns out that changes in Left success in local government elections

were primarily explained by changes in the inflation rate in the region, and in the fortunes of

the Congress party in the national Parliament, in combination with historical incumbency

patterns in local areas, rather than changing socio-economic circumstances of the villages

4

concerned. The identification assumption underlying our analysis is that fluctuations in the

fortunes of the Congress party in national politics were uncorrelated with fluctuations in

preferences of rural voters for land reform in their respective villages. This is plausible given

that changes in national political coalitions were driven largely by events at the national

level, such as the assassinations of Indira Gandhi and Rajiv Gandhi, the rise of the Bharatiya

Janata Party and other regional parties in other parts of the country.4

It turns out that the patterns observed earlier continue to hold despite controlling for

endogeneity bias. So we then address the second problem, by developing a model of two

party competition which nests the standard Downsian and ideology-based models, besides

allowing for political moral hazard (in the form of rent-seeking or administrative effort).5

Interactions between moral hazard and evenness of electoral competition provides an expla-

nation for an inverted-U relationship, in terms of the greater susceptibility of a dominant

political party to moral hazard: a more lop-sided electoral contest (arising from more skewed

preferences amongst voters in favor of one party) translates into lower effort by the dominant

party to secure favor from a majority of poor voters by carrying out land reform.

In the case of the land distribution program, we subsequently find empirical evidence of a

statistically significant inverted-U, as well as the interaction effects predicted by the theory.

Correlations with demographic patterns and the local land distribution are also consistent

with the Downsian theory that voter preferences mattered. In the case of the sharecropper

registration program, the results are qualitatively similar but not as statistically precise.

Both programs additionally show significant spikes in pre-election or election years. These

results are difficult to reconcile with an ideology-based hypothesis. We are thus led to

the conclusion that electoral competitiveness motives predominated, in conjunction with

political moral hazard. The results suggest that evenness of political competition matters

in influencing political will, and therefore strengthening electoral institutions to promote

4Lagged Left share of GP positions provides another instrument for current Left share, controlling for

village fixed effects. This econometric specification is not rejected by the data.5This extends hybrid ideology-competition models of Lindbeck-Weibull (1993) and Dixit-Londregan

(1998) to accommodate moral hazard. Similar predictions would also result from the special interest models

of Baron (1994) and Grossman-Helpman (1996), as shown in an earlier version of this paper.

5

political competition can have benign effects on growth and poverty reduction.

The paper is organized as follows. Section 2 describes the institutional background to the

West Bengal land reforms. Section 3 presents descriptive statistics, and Section 4 describes

the polar hypotheses of ideology and opportunism in the context of a probabilistic voting

model. Section 5 presents censored regressions of land reform with respect to Left control

of local governments which indicate evidence against both theories. Section 6 explores

robustness of these findings with respect to endogeneity of Left control. Section 7 then

presents the hybrid model with political moral hazard, and Section 8 tests this empirically.

Finally, Section 9 concludes. Details of the data sources are described in the Appendix.

2 Historical Background

Following Independence in 1947, land reforms were an important priority for newly elected

governments at both the central and state levels in India. These included abolition of

intermediary landlords (zamindars), redistribution of lands above mandated ceilings, and

regulation of tenancy. Responsibility for agricultural policy was vested in state governments

under the Indian Constitution. Respective states proceeded to enact suitable legislation in

the early 1950s, with encouragement and assistance from the central government.

2.1 Land Redistribution Program

Legislation governing land reform in West Bengal for the period under study is defined by

the second West Bengal Land Reforms Act, passed in 1971. This Act imposed a limit of 5

‘standard’ hectares of irrigated land (equal to 7 hectares of unirrigated land) for a family

of up to five members, plus 12 hectares per additional family member, up to a maximum

of 7 hectares for each family.6 Landowners were required to submit a return (Form 7A)

providing details of the lands in their possession, their family size, and the surplus lands that

they would consequently surrender. Problems of implementation of the new Act however

6One hectare equals two and a half acres.Orchards were allowed 2 standard hectares, and religious and

charitable organizations up to 7 standard hectares (except in suitably deserving cases).

6

soon became evident, arising out of the need to identify the genuine family members of any

given landholder (Appu (1996, p.176)), and nonfiling of returns by an estimated one half of

all landholders.

In 1977, the Left Front came into power in the state, displacing the Indian National

Congress which had formed the government for all but three years since Independence. A

left-wing United Front government had briefly taken over for three tumultuous years in the

late 60s, a period too brief to implement any serious structural changes. Since 1977, the

Left Front has won a majority in all subsequent elections to the state legislature, in marked

contrast to all other Indian states where incumbents have frequently lost office and even

otherwise rarely secure an outright majority.

Immediately upon forming the government, the Left Front set about implementing the

1971 West Bengal Land Reforms Act, which had been amended in 1972. Some of the legal

loopholes (registration of land in religious and charitable trusts, and conversion into ponds)

were sought to be closed by a new amendment in 1981, which however was approved by

the country’s President only in 1986. The President had inserted a number of statutory

provisions (e.g., requiring the government to issue notices to landowners, and wait for their

returns before taking any action to recover surplus lands above the ceiling), which reduced

their effectiveness considerably. Owing both to this and the high degree of fragmentation

of holdings in process for a few decades that reduced the amount of land in large holdings,

the government was not able to vest (i.e., secure land from surplus landholders) as much

land as it had originally hoped to.7

Where the Left Front appears to have distinguished itself was distribution of vested

lands in the form of land titles or pattas to poor households, and the tenancy registration

program of Operation Barga. According to most accounts, this was largely a matter of po-

litical will (see, for example, Appu (1996), Bergmann (1984), Kohli (1987) or Lieten (1992,

p. 128-9)). A massive mass-mobilization campaign involving party leaders, local activists

and the administrators was mounted to identify surplus or barga land, and distribute (or

7Compared to other states, however, the West Bengal government appears to have performed significantly

better on this dimension — whereas in the early 70s West Bengal had an estimated 1.8% of all declared

surplus land in the country, by 1985 this had increased to 16% (Lieten (1992, p. 127)).

7

register) them. Election to local governments (panchayats) were mandated from 1978 on-

wards, and the active cooperation of the newly elected bodies was sought in this process.

The panchayats set up land reform committees (Bhumi Sanskar Sthayee Samiti) with the

state government’s junior land reform officer acting as convener, which was empowered to

form ‘lists’ of surplus lands and of suitable beneficiaries. Settlement camps were set up,

with follow up re-orientation camps, to create an open and participatory process for the

preparation of these lists, in which tenants and poor farmers participated widely (Lieten

(1992, pp 135-136)), Pramanick and Datta (1994, p. 17-18) and Webster (1992, pp. 74-78)).

Most commentators have reviewed the outcomes of this process favorably. P.S. Appu

(1996, Appendix IV.3) estimated the extent of land distributed in West Bengal until 1992 at

6.72% of its operated area, against a national average of 1.34%; only one other state (Jammu

and Kashmir) achieved a higher percentage, with the vast majority of states distributing

less than 1.5% of operated area.

2.2 Sharecropper Registration: Operation Barga

The Left Front government amended the law concerning sharecropping contracts immedi-

ately upon assuming power in 1977. These amendments made sharecropping hereditary,

rendered eviction by landlords a punishable offense, and shifted the onus of proof concerning

identity of the actual tiller on the landlord. The 1981 Amendment Act received Presidential

assent in 1986 restricted the scope of the Act to sharecroppers (bargadars) with less than

10 acres.

The Left Front government subsequently made Operation Barga the centerpiece of its

mass mobilization effort of poorer peasants throughout the state. Membership in the Kisan

Sabha swelled from 1.3 million in 1977-78 to 8.5 million by 1990. While the initial drive

in 1979 was interrupted by the floods, the operation was mounted again in 1980, with the

active cooperation of the panchayati institutions. Over a million bargadars were registered

by 1981, up from 242,000 in 1978 (Lieten (1992, Table 5.1)), and increasing to almost

one and a half million by 1990. Lieten (1992, p. 161) estimates on the basis of different

assumptions concerning the actual number of sharecroppers in the state, that upwards

8

of 80% of all sharecroppers were registered in the state. The government estimates of

sharecropper registration are somewhat lower, of the order of 65% by the early 1990s.

3 Descriptive Statistics

Table 1 provides the district-wise breakdown of the sample, as well as the percent seats

in the GPs secured by the Left front alliance party. The Left secured a majority in most

districts. The mean proportion of GP seats secured by the Left was 69%, with the median

slightly higher, and with the first quartile at approximately 50%. In three quarters of the

GP administrations, thus, the Left obtained an absolute majority. Most of the variation

in the political composition of the GPs thus represented the extent of majority control

of the Left, rather than switches in majority between the Left and its principal rival, the

Indian Congress party (or Congress, for short). Table 2 shows variation in the control and

popularity of these two parties over successive elections, aggregated across all the sample

villages. The control of the Left waxed somewhat in the 1983 elections, falling from 74% to

63%, but recovered thereafter to between 68–70% of GP seats. The vote shares of the Left

and Congress in the immediately preceding elections to the state legislature (or Assembly,

for short) indicate that these two parties together accounted for over 80% of all votes cast.

Their collective share of all GP seats was above 90% on average. Accordingly we shall view

the political contest as involving just these two parties.

Table 3 provides economic and demographic characteristics of the sample villages and

how these changed between 1978 and 1998. There was a sharp increase in the number of

households within villages, owing to population growth, migration, and splits of joint house-

holds. Landlessness grew, with landless households comprising over half the population by

1998. On the other hand, among landowners the land distribution became more equal, with

a significant redistribution of land (about 20%) from medium and big holdings to marginal

holdings.8 Illiteracy rates fell, especially among the landless and small landowners. Wage

rates rose by 28%, nonagricultural occupations became more important, and farm yields

8These shares closely parallel shares of different size classes in the state Agricultural Censuses between

1980 and 1995, lending confidence to our survey estimates of the land distribution.

9

doubled between the early 1980s and mid 1990s.

Tables 4 and 5 provide details of the land redistribution program. Between 15–20%

of operational land area had been vested, or secured from surplus owners by 1998. This is

consistent with the estimate reported by Appu (1992). However most of the vesting occurred

prior to 1978, confirming accounts that the Left Front did not achieve much progress on this

dimension since coming to power in 1977. Their achievement was much greater with regard

to distribution of land titles or pattas to the landless, as shown in Table 5. Approximately

70–75% all patta land in 1998 had been distributed after 1978. Most of the distributed land

was cultivable (ranging between 70 and 90%). We shall therefore focus on patta distribution

rather than vesting operations when examining the land redistribution program.

Distributed patta land in our sample constituted about 3% of operational land area in

the Gangetic part of West Bengal, and 5.5% for the state as a whole, somewhat below the

state government’s own statistics or the estimate of Appu already cited. The proportion

of households receiving pattas was 15%, higher than the proportion of operational land

area distributed. Patta holders constituted about 30% of all landless households, roughly

consistent with the statistics quoted by Lieten (1992). The land distribution program was

therefore far more significant in terms of the number of households that benefited from the

program, rather than actual land area distributed. Most recipients received plots below 1

acre in size, substantially below average holding sizes in the village.

The fact that land distributed (3–5%) was substantially less than the total amount

of land vested (16%) is somewhat surprising. One typically expects appropriation rather

than distribution to be the difficult component of land reform implementation, from either

political, legal or administrative standpoints. Why wasn’t the government distributing lands

it had already appropriated? One can only surmise the reasons for this, based on anecdotes

and opinions expressed by various people associated with the reforms. One account is that

lands officially listed as vested were still under litigation, and the process of identifying

suitable beneficiaries and granting them official land titles was lengthy and cumbersome.

Another is that local landed elites exercise influence over local governments to prevent

distribution of land titles to the poor, for fear that this will raise wage rates of hired labor,

10

and reduce dependence of the poor on them for credit and marketing facilities. The most

common account is that elected officials exploit undistributed vested lands for their personal

benefit in various ways. For instance, informal accounts allege that undistributed vested

lands are used by GP officials to allocate to select beneficiaries to cultivate on a temporary

basis, as instruments of extending their political patronage. There may also be outright

corruption whereby GP officials extract rents from the assigned cultivators.9 Irrespective

of which is the correct story, it is evident that the the availability of vested land did not

constrain the distribution of land titles; instead political will did.

Equally surprising is how small the patta program was in comparison to the changes

in land distribution occurring through market sales and/or household subdivision. Recall

from Table 3 that the proportion of non-patta land in medium and big holdings declined by

about 20%, through land sales or subdivision, and fragmentation of landholdings resulting

from splitting of households. This ‘market’ process was thus almost six times as large as

the redistribution achieved by the patta program, and thus unlikely to have been ‘caused’

by the latter. Accordingly we shall use the distribution of non-patta land as an independent

determinant of voter demand for land reform.

Turning now to the sharecropper registration process, Table 6 shows that the proportion

of cultivable land affected was of the order of 6–7%, and the proportion of households

registered was approximately 3–5%. Hence the barga program represented approximately

twice the land area, but considerably fewer households (one-third), compared with the patta

program. The proportion of sharecroppers registered was of the order of 50%, slightly below

the state-wide registration rates of 65% reported in the early 1990s.

Regarding timing of the reforms, the bulk occurred in the first two local government

administrations spanning the ten year period between 1978 and 1988. This is shown in

Table 7. Table 8 shows that a significant fraction of villages in the sample witnessed no

reforms at all — over one quarter over the entire twenty year period. Only 4–6 villages

out of 88 witnessed some reforms in every single administration. This indicates the need to

9We have been informed of this by Debu Bandyopadhyaya, the Land Reforms Commissioner during the

late 1970s and early 1980s. We have also recently heard such accounts in the course of our currently ongoing

surveys of these villages.

11

incorporate endogenous censoring in the empirical analysis.

4 Theory: the Polar Hypotheses of Ideology and Oppor-

tunism

In this section we present a model of probabilistic voting with electoral competition be-

tween two parties with differing assumptions concerning objectives of these parties, which

correspond to polar hypotheses of ideology and opportunism.

Consider any village v with total voter population normalized to unity, where voters

belong to different landowning classes c = l, g, s, m, b consisting respectively of the landless,

marginal,small, medium and big landowners. The last category consists of those holding

land above the legislated ceiling, from whom the government may seek to vest lands and

distribute to the landless. The demographic weight of class c is αc. Elected governments

select a policy π from some policy space P . Preferences of a voter in class c are represented

by utility Uc(π).

There are two parties denoted L and R. Let the policy of a party p candidate or elected

official be denoted πp. These represent either the policy platform of the candidate prior

to the election, which the candidate is committed to in the event of being elected. Or it

represents the policy actually carried out by the candidate while currently in office. In this

case, we shall assume that voters project the current policies into their future expectations,

so voting behavior in the next election is determined by these policies.

A fraction τc of class c voters turn out to vote in the election. Of these, a further

fraction βc are aware voters, the rest are impressionable.10 Aware voters respond to policy

differences as well as intrinsic loyalties to the two parties which are exogenously determined.

Impressionable voters vote entirely on the basis of their loyalties. We assume that within

village v, relative voter loyalty to the party L candidate is distributed uniformly with density

fc (which may be specific to the class c the voter belongs to) and mean εdct.

10Grossman and Helpman refer to them as ‘informed’ and ‘uninformed’ in their 1996 article, and as

‘strategic’ and ‘impressionable’ in their 2001 book.

12

An aware voter in class c with loyalty ε votes for the L party candidate if Uc(πL) + ε >

Uc(πR). An impressionable voter with relative loyalty ε to the Left party votes for that

party as long as ε > 0.

The resulting vote share of the Left party is (with γc denoting τcfc:)

VL =12

+1

∑c αcτc

[∑

c

αcγcεdct +

∑

c

αcβcγc{Uc(πL) − Uc(πR)}]. (1)

Vote shares determine the probability φL of the Left party winning the election, accord-

ing to φL = φ(VL), a strictly increasing, continuously differentiable function. The presence

of randomness in election turnout, and errors in vote counting cause this function to be

smooth rather than a 0 − 1 discontinuous function.

Turn now to the objectives of parties. In the Downsian model, each party has no

intrinsic policy preferences, and seeks only to maximize the probability of being elected.

Then each party seeks to maximize its vote share. It follows from expression (3) that

both parties must select the same policy π∗ which maximizes∑

c αc[βcγc]Uc(π)], a quasi-

utilitarian welfare function in which the welfare weight for any class βcγc is the product of its

turnout (τc), voter awareness (βc) rates, and the proportion of swing voters (fc) within that

class. In this case both parties converge to the same policy that is desired by the ‘average’

informed swing voter — implying that party background is irrelevant in predicting policy

choices. The probability that any given party is elected is then determined entirely by the

(exogenous) voter loyalties. Note that this result does not necessarily require that parties

or candidates commit to their policies in advance of the elections. It could be the result of

choices actually made by elected officials of either party, with the objective of maximizing

the probability of being subsequently re-elected.

In the ideology model, parties have intrinsic preferences over policy choices. These could

be ideologically determined in part, or driven by the interests of constituents that the party

represents. For expositional convenience, however, we shall refer to these as ‘ideology’, and

represent it by a set of welfare weights wic assigned by party i to the interests of class c. It

is natural to suppose that the Left party assigns greater weight to classes owning less land,

with the opposite true for the Right party, so the ideologically desired policies by the two

13

parties are ordered, with the Left party desiring greater land reform: π∗L > π∗

R where π∗i

maximizes∑

c αcwicUc(π).

In addition, one needs to assume that elected officials neglect the effect of current policies

on future re-election. Then elected officials select their ideologically most preferred policies.

Votes are cast in elections where voters anticipate these policy choices, so the probability of

winning the election are determined partly by ideology or policy differences, and partly by

voter loyalties. In contrast to the Downsian theory, here policies are determined by party

rather than voter preferences, and policy convergence does not occur. The more Left party

candidates in office, the greater will be the extent of land reform.11 Exogenous swings in

voter loyalties in favor of the Left will raise the fraction of Left candidates elected, and

leave the policies espoused by either party unchanged, resulting in greater land reform.12

In contrast, the Downsian model predicts that there will be no effect on land reform. Hence

the two models generate distinct predictions for the correlation between land reform and

the fraction of GP positions secured by the Left.

5 Preliminary Empirical Patterns

Tables 9 and 10 present regressions of different measures of land reforms implemented with

respect to the Left share of GP seats. The different measures are pattaland: proportion

of cultivable land in the village distributed in the form of pattas; pattadar: proportion of

households who received pattas; bargaland: proportion of cultivable land registered under

the barga program; and bargadar: proportion of households registered in this program.13

11Each local government or GP is a council of elected members, one from each electoral constituency. It

is reasonable to suppose that the chosen policy weights the desired policies of both parties, with the weight

on any party increasing in the fraction of seats secured by that party.12Indeed, given the natural assumption that party ideologies do not change from one election to the next,

changes in relative electoral success of the two parties must result from such loyalty swings, so there is a

one-to-one relation between these two variables.13We do not use the barga registration rate owing to the significant underreporting of tenancy in the

household survey, which artificially inflates the registration rate. For almost forty villages no land was

reported as under tenancy in 1978, while significant numbers of bargadars were recorded in the 1970s in

14

Owing to the significant censoring in the data, we report results of tobit regressions.

The cross-section tobits aggregate across the entire twenty year period 1978–98, while the

panel tobits aggregate within each five year period spanning a single GP administration14,

and use dummies for districts as well as for the four time blocks. We do not use village

fixed effects because of the well known inconsistency of tobit estimators with village fixed

effects. The number of fixed effects to be estimated declines substantially when they are

at the level of the district: consistency of the estimator refers to limiting properties as the

number of villages per district grows large, assuming that all the unobserved heterogeneity

arises at the district rather than village level. The cross-section patta tobits controls for the

proportion of land vested by 1978 which represented the land available for distribution, and

the population density in 1978 which represents a measure of the demographic pressure for

land distribution. The cross-section barga regressions control for the extent of unregistered

barga land or households in 1978, which represented the potential for registration.

In no case do we see evidence of a significant relationship of land reforms implemented

with Left control of the local GP. The signs and magnitudes of the estimated coefficients

imply that the nature of this relationship follows an inverted-U, with a turning point at or

below the mean and median Left share. This implies that for the majority of the sample,

higher Left control was associated, if at all, with less land reform. This is clearly contrary

to the pure ideology model. The absence of a statistically significant relationship may be

viewed as consistent with the predictions of the Downsian model.

However, the simple correlation between land reform and Left control may conceal

the true underlying relationship, if there are other factors correlated jointly with both

outcomes. For instance, it is natural to expect the landless to have a greater demand for

land reform, and greater loyalty to the Left. A rise in landlessness would then increase

both variables under either theory, implying that the observed correlation overstates the

those villages. The registration rate cannot be computed for these villages. We therefore express the scale

of the barga program relative to the total cultivable area and number of households in the village instead.14Election years are treated as part of the time block of the outgoing administration, given the existence

of lags arising from legal delays and the fact that a new administration usually assumes office in the second

half of the year.

15

true (partial) correlation. This is when turnout, awareness and ‘swing’ factors among the

landless are not significantly lower than for the remaining population. If on the other hand

the landless are significantly less aware, turn out less or less prone to switch votes in response

to policy differences than other classes, then a rise in demographic weight of the landless

would generate less land reform under the Downsian theory. Then rising landlessness would

generate a negative correlation between land reform and Left control, under the Downsian

theory. To test this theory, it is therefore important to control for demographic proportions

of different size classes, or other village characteristics that might be correlated with either

land reform or Left electoral success.

Tables 11 and 12 present results for censored regressions which control for village char-

acteristics such as demographic weight of different landowning size classes, land shares,

proportion of households belonging to scheduled castes or tribes, and illiteracy rates of non-

big (which aggregates landless, marginal and small landowners) and big landowners. Land

size classes are defined by ownership of cultivable non-patta land, with marginal, small,

medium and large categories defined by 0-2.5, 2.5–5. 5–12.5 and 12.5– acres respectively.

Village demographics, land distribution and illiteracy rates are interpolated for different

time blocks based on their respective rates of growth between the two survey years.

Tables 11 and 12 also reports results for the semiparametric trimmed LAD estimator

with village fixed effects proposed by Honore (1992). Besides controlling for inter-village

heterogeneity and censoring, the latter estimator avoids the normality assumption on resid-

uals, replacing it with only a symmetry (i.i.d.) restriction on the distribution of residuals.

Moreover, Tables 11 and 12 report regressions applied to yearly data, instead of aggre-

gates across five-year timeblocks corresponding to different elected GPs, and controls for

election and pre-election year dummies. Use of yearly data permits controlling for the tim-

ing of elections, which is an additional way to discriminate between hypotheses of electoral

opportunism and ideology. The former model would predict higher land reforms immedi-

ately before an election, when voters are more likely to project current policies to the future,

and elected officials pay more attention to their re-election prospects. Such election year

effects should not arise in a pure ideology model.

16

In addition, the use of yearly data permits the coefficients to be more precisely estimated

owing to an expansion in sample size, compared with data aggregated into five year time-

blocks corresponding to the tenure of each elected GP. On the other hand, standard errors

may be underestimated with yearly data if residuals are highly correlated across different

years within a given GP administration in a given village. In that case observing reforms

implemented in different years within the tenure of a given GP does not really constitute

more information than is contained in the aggregate for the entire five year block. How-

ever, this is a problem which tends to be significant only if the number of ‘groups’ is small

(Donald and Lang (2004)): in our context here a ‘group’ corresponds to a village-timeblock

combination, and our dataset contains over 300 village-timeblocks. So this problem should

not be particularly significant. Nevertheless the presence of serial correlation may imply

that the standard formulae for standard errors are subject to bias, if the nature of this serial

correlation cannot be represented by a fixed shock for each village-timeblock combination,

as pointed out by Bertrand, Duflo and Mullainathan (2004). For a later specification of

the TLAD regression we shall examine the robustness of our results with respect to this

potential problem, by reporting standard errors obtained from bootstrapping the estimator.

We see that virtually the same patterns as in Tables 9 and 10 are repeated in Tables 11

and 12 — absence of a monotone increasing relationship with Left share of GP seats, and an

inverted-U pattern whenever the coefficients are statistically significant (which is the case

for the pattadar and bargadar regressions with village fixed effects). There is a significant

election year positive spike in barga activity, and a negative election year spike in patta

activity. GP elections are held in the April of an election year, and the new government

tends to take charge by June. So an election year represents three months of an outgoing

administration, and six months of an incoming one. Both a positive or negative spike is

then consistent with the Downsian theory, where the positive component results from the

higher than average efforts of the outgoing administration or its hangover, and a negative

component from the lower than average efforts of the incoming one. It is much harder to

rationalize them under the ideology hypothesis, since party ideologies are unlikely to change

from one year to the next. So the evidence continues to still be inconsistent with the ideology

hypothesis. The inconsistency with the Downsian hypothesis is substantially less, but the

17

problem of explaining a significant inverted-U in pattadar and bargadar regressions remain.

It is possible, of course, that unobserved components of village voter preferences that

simultaneously change land reform and Left electoral success account for an inverted-U

under either hypothesis. The only way to deal with this problem of potential endogeneity

of Left control, is to seek suitable instruments for this variable.

6 Predicting Success of the Left in Local Elections

Probabilistic voting models allow voting behavior to reflect both loyalties of voters to dif-

ferent parties for various exogenous reasons (such as historical factors, incumbency, the

specific characteristics of candidates etc.), as well as their policy preferences. We can there-

fore search for measures or determinants of voter loyalty to the Left that reflect factors

external to the village, or historical circumstances orthogonal to land reform related issues

in the current election. The Left and Congress contest elections at different levels, such as

the state and federal legislatures (which we shall henceforth refer to as the Assembly and

Parliament). These elections are staggered across different years: the Assembly elections

are typically held one or two years before the GP elections (they were held in 1977, 1982,

1987, 1991 and 1996). The Left and the Congress are the principal adversaries in the state

assembly elections, as well as elections for seats representing West Bengal constituencies in

Parliament.

Given that local government elections were introduced for the first time in 1978, and

that most voters in India tend to view politics in terms of state or national rather than local

issues, it is plausible to suppose that voter loyalties in local elections were determined to a

large extent by regional or national issues. These may include inflation in consumer prices,

growth in factory employment, or the relative strength of the two parties in the national

Parliament. The Congress formed the national government between 1980 and 1984, and

reinforced its position between 1984 and 1989 following the assassination of Indira Gandhi

in 1984. Between 1989 and 1991 a non-Congress government prevailed at the national level,

representing a coalition of different regional parties supported by the West Bengal Left

18

Front. Then again from 1991 until 1996 the Congress formed a government at the national

level, with the Left in the opposition.

The fluctuating strength of the two parties in Parliament had considerable implications

for relations between the central and the state government over fiscal transfers, execution

of central government projects in the state, and other matters likely to have significant

spillovers into inflation, employment and public services in rural areas. For instance, a

Congress party member elected to a Parliamentary seat from a specific West Bengal con-

stituency would be in a position to direct public projects in railways and irrigation to his

constituency, thus boosting loyalties of voters towards the Congress in local GP elections.

Conversely, Left candidates often blame Congress dominated central governments for starv-

ing the state of fiscal transfers or public investments for partisan reasons, and use this in

their election rhetoric in order to mobilize voters against the Congress party.

Table 13 presents regressions predicting Left control of local GPs, on the basis of a

variety of factors both external and internal to the villages in question. The external

factors include the proportion of seats secured by the Congress in the currently elected

Parliament, the rate of inflation over the past five years in price index for agricultural

workers in the nearest regional center (among four centers in the state for which this index

is constructed by the government: Calcutta, Jalpaiguri, Ranigunj and Asansol), and the

growth in factory employment in the district over the last five years. The inflation and

employment variables are constructed at a much higher level of aggregation than a single GP:

there were approximately 200 GPs in each district, and between three and five districts in

each region. Hence these variables reflect economic changes covering a much larger area than

the jurisdiction of a single GP. We also include the average vote share difference between Left

and Congress candidates in the immediately preceding state Assembly elections, averaged

at the district level, as a proxy for prevailing voter loyalty to the two parties on the basis

of district, state or national issues.

The local factors that may affect electoral success of the Left in GP elections include

incumbency patterns in the GP, besides land distribution, literacy and caste in the village.

The regressions interact local incumbency with external factors, since voters reaction to

19

changes in inflation and employment may depend on which party has been dominating in

the local area. For instance, a rise in inflation is likely to cut into support of the local

incumbent and strengthen the position of the opposition party.

Table 13 shows results of the GP Left share regression applied to five successive GP

elections (1978, 1983, 1988, 1993 and 1998). The first two columns show cross-section least

squares results, while the remaining three show the panel estimates. In the panel we use

the Arellano-Bond (1991) estimator to avoid the bias that arises from a lagged dependent

variable (incumbency) as a regressor. The hypothesis of lack of first-order serial correlation

in the level of the time-varying errors (equivalently lack of second-order correlation in the

differenced residuals) is not rejected at most conventional levels of significance. Hence

(controlling for village fixed effects) lagged Left share is also a valid instrument for Left

share in the land reform regression.

The cross-section results show that the assembly vote share difference at the district

level was a strong predictor of local GP outcomes. Of remaining village characteristics,

only caste and literacy among the poor were statistically significant influences. The local

land distribution was not significantly related to local GP control secured by the Left. In

the panel, local literacy or caste become unimportant, while the role of voter loyalty at the

district level remains robust (last column of Table 13). Voter loyalties at the district level,

in turn, are shown in Table 14 to be related to lagged loyalty, regional inflation rate, and

Congress presence in Parliament.

The third and fourth columns of Table 13 replace the district voter loyalty variable

by its underlying determinants directly. These show that changing fortunes of the Left

in GP elections (conditional on incumbency patterns) were driven mainly by changes in

factors external to the villages — the regional inflation rate, and the presence of Congress

in the national Parliament — rather than changes in village characteristics.15 The nature of

these effects are intuitively plausible: rising inflation hurt the local incumbent, while rising

Congress fortunes at the national level helped the Congress in GP elections in constituencies

15Inclusion of lagged land reform in the village concerned on the right hand side did not yield a significant

coefficient either, irrespective of whether it was included by itself or in interaction with incumbency.

20

where they were already strong. This may reflect the tendency for Congress ministers in the

national government to favor their own constituencies in the location of central investment

programs. Somewhat more surprising is that a rise in the presence of the Congress at

the national level also benefited the Left party in areas where the Left was traditionally

powerful. We presume this reflected the ability of the Left to gain mileage in local elections

by blaming a Congress-dominated national government for local problems.

6.1 Instrumental Variable Estimates of the Land Reform-Left GP share

relationship

The preceding results imply that external and historical factors driving the fluctuations

in Left control of local GPs are suitable instruments which can be used to correct for

potential endogeneity of the Left share variable in the land reform regression. The presence

of Congress in the national Parliament seems particularly suitable as an instrument, as

these principally reflected the increasing importance of coalition politics at the national

level, and other events in the rest of the country (seccessionist movements in Punjab and

Kashmir, the assassinations of Indira Gandhi and Rajiv Gandhi which subsequently created

a pro-Congress wave, rising power of regional parties and the Bharatiya Janata Party in

other parts of India, and border tensions with Pakistan)). These factors are likely to be

uncorrelated with time-varying village specific voter preferences for land reform. Moreover,

Table 13 shows we cannot reject the hypothesis of absence of serial correlation in the Left

share regression after controlling for village fixed effects, which implies that incumbency

(lagged Left share) is also a valid instrument.

Accordingly, we use the fourth column of Table 13 to predict the Left share in each

GP, and then use these to generate instrumental variable tobit and TLAD estimates of the

land reform-Left share regressions. The correlation between the predicted and observed

changes was .32, so the IV estimates are unlikely to suffer from a weak instrument problem.

The IV estimates of the tobit and TLAD regression coefficients of Left share of GPs and

election/pre-election year dummies are reported in Tables 15 and 16. The standard errors

do not correct for the prediction errors with respect to the first stage regression, so represent

21

underestimates of the true standard errors. This does not matter here because even with

the underestimated standard errors we see no evidence of a significant positive relationship

between local Left control and land reforms implemented. Only in the case of the village

fixed effect TLAD regression for bargadar do we see a statistically significant relationship

at 5% significance or less, but this relationship follows an inverted-U. The election year

dummy is significant in this regression. Thus we continue to not find evidence in favor of

the pure ideology hypothesis.

The weak relationship with local Left control evidence indicates closer support for the

Downsian hypothesis. However, the inverted-U pattern for bargadar is not consistent with

it. It makes sense, therefore, to probe possible reasons for an inverted-U relationship. This

is the topic of the next section.

7 A Hybrid Model with Moral Hazard

We argue in this section that the inverted-U pattern can be interpreted as reflecting a form

of moral hazard among elected politicians, wherein those in a strong competitive position

(vis-a-vis potential challengers in the next election) can afford to slacken their land reform

effort. Moral hazard can emanate from costly effort that elected officials must undertake

in order to implement reforms, or from rents foregone from land vested that has not been

appropriated. A conflict of interest between voters and elected officials over the desired

extent of land reform can then emerge. Higher levels of land reform will be observed when

the electoral contest is more even, with the GP composition closer to 50-50. We now develop

an extension of the previous model with probabilistic voting to incorporate moral hazard.

We also extend the model to incorporate costly election campaigns mounted by parties

to mobilize impressionistic voters. Apart from moving the model closer to reality, it provides

parties with an additional instrument of electoral strategy which will turn out useful in the

characterization of electoral outcomes. Let Mi denote the size of the campaign mounted

by party i at cost θiMi, where θi > 0 is a given parameter. Then as in Grossman-Helpman

(1996) we assume that (while aware voters vote as supposed previously) an impressionable

22

voter with relative loyalty ε to the Left party votes for that party as long as h[ML−MR]+ε >

0, where h > 0 is a given parameter. The resulting vote share of the Left party is then

12

+1

∑c αcτc

[∑

c

αcγcεdct +

∑

c

αcβcγc{Uc(πL) − Uc(πR)}

+ h∑

c

αcγc(1 − βc)(ML − MR)]. (2)

Denote by χ ≡ h∑

c′ αc′τc′(1 − βc′)fc′ a parameter which represents the value of electoral

campaigns in mobilizing voters, which is proportional to the fraction of impressionable

voters. Then the vote share expression can be simplified to

VL =12

+1

∑c αcτc

[∑

c

γcεdct +

∑

c

αcβcγc{Uc(πL) − Uc(πR)} + χ(ML − MR)]. (3)

In contrast to the Grossman-Helpman (1996) theory, we assume that campaigns are financed

by parties themselves, rather than from contributions raised from special interest groups.

The previous version of this paper shows that similar results obtain in the presence of

campaigns financed by special interests such as big landowners that resist the reforms.16

Moral hazard arises from private costs to elected officials (either effort or foregone rents)

that depend upon the extent of land reform: e = e(π). Party objectives represent a mixture

of opportunism, ideology and moral hazard. The opportunistic component stems from the

opportunity to earn rents while in office. Part of these rents are exogenously fixed, and

denoted Ei for party i. These could represent ‘ego-rents’, or pecuniary rents arising from

the power of officials over other areas of policy apart from land reform. The remaining

variable rent component is represented by −ei(π), and hence the total rent is Ei − ei(π).

The combination of ideology and rent-seeking implies that the objective of candidates

is to maximize the expected value of

∑

c

αcwicUc(π) − e(π). (4)

The ex ante payoff of party i (with j �= i) denoting the other party, and φi, φj ≡ 1−φi their

16We choose the formulation with internal financing and political moral hazard because this seems more

natural in the West Bengal context. The special interest model would be driven by capture of the Left party

by big landowners, which seems rather far fetched.

23

respective win probabilities, is then

Oi(πi, Mi; πj , Mj) = φi[∑

c

αcwicUc(πi) − ei(πi) + Ei]

+(1 − φi)∑

c

αcwicUc(πj) − θiMi. (5)

This formulation presumes that parties commit to policy platforms in advance of the

election. The same characterization of equilibrium policy choices holds when such commit-

ment is not possible, but with voters forecasting future policies from current ones, so the

vote shares in the next election are given by the same function (3) of current policy choices.

Let Di denote the expected rents from future office, and δi the discount factor of a party i

incumbent. Then this incumbent will select πi, Mi to maximize

∑

c

αcwicUc(πi) − ei(πi) + Ei − θiMi + δiφi(Vi)Di. (6)

This model nests different polar theories of political competition. The Downsian model

obtains when we assume that candidates have no ideological preferences (wic ≡ 0), nor any

policy-related sources of personal rents (ei(πi) ≡ 0).17 The pure ideology model obtains

when incumbents cannot commit to their future policies, earn no rents (Ei = ei ≡ 0), and

discount the future at a high enough rate that they ignore implications of current policy

choices on future re-election prospects (δi ≡ 0).

The more general version presented here admits a hybrid of electoral opportunism, rent-

seeking, and ideology. The ingredients we add to the model can all be justified by an appeal

to the reality of the West Bengal political context, besides the need to accommodate the

facts. It is well known that the Left parties have been subject to internal debate concerning

the need to strike a balance between its traditional ideology and opportunism.18 As a

17Then with commitment the payoff of i reduces to maximization of φiEi−θiMi, and with no commitment

reduces to maximization of δiφiEi − θiMi. Hence the policy πi chosen by i must maximize the probability

of winning φi. Expression (3) shows that both parties will select the same policy π∗ which maximizes∑

cαcβcγcUc(π).

18See, e.g., Franda (1971), Chatterjee (1984), Nossiter (1988), Lieten (1992, pp.128-133) and Bhattacharya

(1999)). The transition of the CPI(M) from a revolutionary party in the 1940s to subsequent capture and

consolidation of the state government is generally attributed to the pragmatism of its leaders Jyoti Basu

24

reading of Lindbeck-Weibull (1993) and Dixit-Londregan (1998) indicates, such a model is

quite complex, and it is not evident from their results whether such a model can account for

the empirical findings reported in the previous Section. That is the question we now pose.

The following proposition represents the main prediction of the hybrid model concerning

equilibrium policy choices.

Proposition 1 Consider any equilibrium of the hybrid ideology-competition model (either

with or without policy commitment) in which both parties select positive campaign levels,

voter utilities are differentiable, and the policy space is an open interval of a Euclidean

space. The policy choice π∗i of party i will maximize

∑

c

αcµicUc(π) − ei(π) (7)

where the welfare weights are given by

µic = wi

c +θi

χ.φ∗i

τcβc

fc(8)

and φ∗i denotes the equilibrium probability of party i winning.

Proof of Proposition 1: Consider the version with policy commitment, where the payoffs

are given by (5); an analogous argument applies in the no-commitment case (with payoffs

(6)). Note that the payoff of party i can be written as φ(Vi)Di +∑

c αcwicUc(πj) − θiMi,

where Di ≡ ∑c αcw

ic{Uc(π∗

i ) − Uc(π∗j )} − ei(π∗

i ) + Ei denotes the winning stakes for party

i. The first order condition with respect to choice of campaign level Mi yields φ′iDiχ = θi.

The first order condition for policy choice yields

φ′iDi∑

c αcτc

∑

c

αcτcβc

fc

∂Uc

∂πi

+ φ∗i [

∑

c

αcwic

∂Uc

∂πi− e

′i(πi)] = 0 (9)

and Promode Dasgupta who consciously chose an approach that would secure widespread political support

with voters, at the cost of disenchantment of some of the party’s ideologues. Lieten provides some of the

internal critiques of the Left Front government’s performance from those disillusioned with its compromise

with traditional ideology. Bhattacharya describes the political transition of the CPI(M) in West Bengal as

pursuing the ‘politics of middleness’.

25

Using the property that φ′iDi = θi

χ , the first order condition for the policy choice can be

written as∑

c

αc[wicφ

∗i +

θi

χ

τcβc

fc]∂Uc

∂πi= φ∗

i e′i(πi) (10)

from which the result follows.

Equilibrium winning probabilities φ∗i will depend in turn on chosen policies, election

campaigns and patterns of voter loyalties, as represented by the expression for vote shares

(3). These are jointly determined along with equilibrium policies and campaign sizes. Nev-

ertheless, equilibrium policy choices π∗i have the property that they maximize

∑

c

αcµicUc(πi) − ei(πi), (11)

a mixture of ideological, opportunistic and rent-seeking motives. Expression (8) for the

implicit welfare weight µic on interests of class c voters indicates that the importance of

the opportunistic motive relative to ideology and rent-seeking depends negatively on the

equilibrium winning probability. A ceteris paribus shift in voter loyalty to party i will raise

its equilibrium win probability, inducing a lower weight on the opportunistic component.

If parties’ ideological motivations are relatively mild (wic’s are small and close to zero), the

effect will be higher levels of rent-seeking. Since higher land reforms are associated with

greater effort, and/or a sacrifice of personal benefits from undistributed vested land, this

implies a cutback in the extent of land reform. In this case we would observe a monotone

decreasing relationship between the share of party i in GP seats (which reflects φ∗i ) and the

extent of land reform π∗i pursued by party i officials. This represents a formalization of the

old adage that ‘power corrupts’.

What are the implications for observed land reform outcomes? It is reasonable to assume

that the actual policy pursued by a GP is a compromise between the policies pursued by

candidates of the two respective parties, with weights that reflect the strengths of the two

parties in the GP. Let the weight on party i’s policy choice be denoted by qi = q(φ∗i ), an

increasing function of party i’s share of GP seats, lying between 0 and 1, and with the

property that q(0) = 0, q(1) = 1. Specifically, the observed land reform outcome π∗ in the

26

GP is given by

π∗ = q(φ∗i )π

∗i + [1 − q(φ∗

i )]π∗j . (12)

Let πIi denote the ideal policy for party i which it would pursue in the absence of any

opportunistic motive, i.e., which maximizes∑

c αcwicUc(π) − e(π). And let π∗ denote the

Downsian equilibrium policy outcome, which maximizes∑

c αcβcγcUc(π). Note that the

Downsian policy does not incorporate the personal rents of elected officials. If the extent

of land reform π is a one dimensional variable then for reasons explained above one would

expect e(π) to be an increasing function. Then the extent of land reform will tend to be

underprovided as a result of the political moral hazard problem. This will be mitigated

only if party i has a sufficient ideological preference for the reform.

Consider the case where the political moral hazard problem dominates ideological con-

siderations, in the sense that the Downsian policy π∗ strictly exceeds the ideal policy πIi of

both parties i = L, R. This is illustrated in Figure 1. Call this Case 1 from now on. Here a

rise in its win probability causes the equilibrium policy of the Left to move closer to its own

desired policy πIL, i.e., it carries out less land reform. At the same time the Right party

will implement more land reform in order to recover ground with voters. If the Left party

was carrying out more land reform initially, the gap between the two parties will narrow.

As voters continue to shift loyalty to the Left party, eventually the gap will vanish and then

get reversed, with the Right party carrying out more land reform than the Left.19

The comparative static effect is different in the other case (referred to as Case 2) where

the moral hazard effect is weaker than the ideological effect so that the Left party intrin-

sically desires more redistribution than the Downsian mean voter (πIL > πD). Suppose

that the Right party still desires less redistribution than the Downsian outcome. Then an

increase in its win probability motivates the Left party to carry out more redistribution.

The Right party also wishes to carry out more redistribution. In this case both parties

carry out more land reform with a shift in voter loyalty to the Left, as illustrated in Figure

19The reason is that (by virtue of Proposition 1) as the win probability of the Left party approaches one,

the equilibrium policy of the Left party will approach its own desired level πIL while the Right party will

approach the Downsian policy πD. Hence there will exist some value of φ∗L where their respective policies

will cross.

27

2. Moreover here the Left party will always carry out more redistribution than the Right

party (since the Left will always want to carry out more than the Downsian policy, and the

Right party less than the Downsian policy). In this case the results will be akin to the pure

ideology model: there will be a monotone, increasing relationship between Left share of GP

seats and the extent of land reform.

8 Testing the Hybrid Model: Methodology and Results

We now describe the regression specification implied by the model described above. Recall

equation (12) which generates the land reform outcome in any village as a function of the

Left share of the GP, and the policies pursued by the two parties. For village v in year t:

πvt = q(LSvt)(πLvt − πR

vt) + πRvt (13)

where we use a quadratic formulation q(l) ≡ al + bl2, l ∈ [0, 1] for Left control.

The Left share of GP seats is jointly determined along with the policies chosen by

the two parties, besides determinants of voter loyalties. Shifting voter loyalties also affect

equilibrium policy choices in the model, by affecting relative competitive strength of the two

parties. Equilibrium policies also depend on village characteristics affecting the demand for

land reform from voters, as well as the ideologically desired reforms by each party. Let LDvt

denote voter loyalty to the Left relative to the Congress, and Svt a vector of distributional

characteristics pertaining to land, literacy and caste in village v in year t. The policy of the

Right party is then given by

πRvt = λ0 + λ1LDvt + λ2Svt + ηR

vt (14)

and the divergence in policies between the two parties by

πLvt − πR

vt = µ0 + µ1LDvt + µ2Svt + ηdvt, (15)

where ηRvt, η

dvt denote regression residuals.

28

Combining the policy equations with (13), we obtain the following prediction for land

reform:

πvt = λ0 + λ1LDvt + λ2Svt +

+ µ0q(LSvt) + µ1LDvt ∗ q(LSvt) + µ2Svt ∗ q(LSvt) + ηvt (16)

The coefficient µ1 represents the interaction between moral hazard and competition missing

in the pure Downsian and ideology models. The Downsian model predicts no policy diver-

gence (µ0 = µ1 = µ2 = 0) and irrelevance of voter loyalties (λ1 = 0). The pure ideology

model also implies irrelevance of voter loyalties (µ1 = λ1 = 0), while policy divergence is

predicted (µ0 �= 0, µ2 �= 0). The hybrid model predicts that voter loyalties matter for policy.

If political moral hazard is severe enough in the sense explained in the previous section,

λ1 > 0, µ1 < 0.

Note that in the presence of significant interactions between moral hazard and competi-

tion, the land reform regression estimated previously was misspecified. The interaction ef-

fects are correlated with the Left share variable, causing the estimated coefficient of q(LSvt)

to be biased. The sign of this bias depends on the sign of the interaction effect. If Case 1

applies, the moral hazard-competition interaction causes policy divergence to narrow and

get reversed when voters shift loyalty to the Left, causing a downward bias in the estimated

coefficient µ0.

8.1 Empirical Results

Tables 17 and 18 present estimates of the TLAD land reform regressions corresponding to

(16) with village fixed effects. Both versions with observed and predicted Left shares are

shown (the latter using regional and national determinants of voter loyalties as instruments).

The new terms added are the relative loyalty at the district level (district-average vote share

difference in the most recent Assembly elections) as a measure of competitive strength of

the Left, and its interaction with Left share (both linear and quadratic terms). Standard

errors for the IV regressions are obtained by bootstrapping.20 As before, the regressions20To guard against the possibility of serial correlation in the residuals for any given village not captured

by village fixed effects, we use a block bootstrap as recommended by Bertrand, Duflo and Mullainathan

29

control for non-patta land distribution, literacy and low caste proportions, besides dummies

for different timeblocks, villages and election/pre-election year.

In the pattadar regression, we find a significant inverted-U with respect to Left share

itself, and a negative, significant interaction between Left share and relative voter loyalty.

Similar patterns emerge in the pattaland regression but the estimated coefficients are not

statistically significant. Recall also that the model imposed the restriction that the same

quadratic in LSvt applies in levels and in the interaction with voter loyalty, i.e., the associ-

ated turning points of the level and interaction effects are the same. In the IV regressions

this prediction is almost exactly borne out. Finally, the election year effect continues to be

negative and significant at 10%.

Table 18 shows corresponding estimates for the barga program. Here the interaction

effect is statistically insignificant, but has a negative sign in both IV regressions, again with

similar turning point as the Left share variable itself.

We therefore find some evidence consistent with the moral hazard hypothesis: shifts in

voter loyalty in favor of the Left were associated with a reduction in the gap between the

land reform efforts of the Left and the Congress. This provides part of the explanation for

the inverted-U pattern of land reform with respect to Left share. Since there continues to be

an inverted-U with respect to Left share despite controlling for this interaction effect, this

is only part of the explanation. According to the theory the relation with respect to Left

share ought to have been monotone, representing the effect of greater Left control per se over

GP decisions (the function q). The fact that we continue to find a non-monotone relation

indicates that there are still unobserved components of voter loyalty not captured by our

AVSD measure, which continues to bias downward the estimated slope with respect to Left

share. An alternative explanation is that the distribution of voter loyalty is thin towards

the tails (rather than uniform as supposed in the model), in which case the proportion of

marginal ‘swing’ voters declines when voter loyalties become more lopsided to favor the

(2004). All the data for a given village is kept as a single block, and 200 samples are generated by sampling

with replacement from the observed data blocks. Both first stage and second stage regressions are run for

each sample, so that the bootstrapped standard errors incorporate both first stage prediction errors, as well

as serially correlated residuals.

30

Left, causing a decline in land reform implemented by the Left.

The corresponding coefficients of other village characteristics in the patta and barga

regressions are shown in Table 19. The proportion of households receiving land titles was

significantly higher when there were more marginal landowning households, fewer medium

landowners, and a smaller fraction of non-patta land was in small holdings below 5 acres.

This is consistent with the notion that the patta program was responsive to voter preferences:

marginal and medium landowners are likely to have been the most politically active groups

on either side of the land reform issue within villages. Moreover, the greater the inequality

in landownership, the greater the scale of the patta program. Somewhat surprisingly, the

scale (in terms of land area, though not proportion of households) was significantly smaller

when the demographic weight of the low caste households increased, which could owe to

lower rates of turnout, political awareness or swing voters within these groups. No particular

patterns are evident for the barga program, except a similar pattern for land areas involved

to be smaller in the presence of a larger low caste population, and a positive effect of

illiteracy rates among the poor, which is not easily interpreted.

8.2 Extension to pre-1978 Period

The preceding results did not make full use of the land reform data available to us, which

covers 1971–98. It turns out that a significant amount of land reform was carried out even

preceding the advent of the Left Front government in 1977, comparable in magnitude to

the reforms enacted in the years following 1978. This was a period of rising competitive

pressure on the Congress administration at the level of the state government arising from

the increasing political strength of the Left parties during the 1970s. However the pre-1978

period did not have any elected local government, which was the reason that it was excluded

in the previous regressions. The responsibility for implementing land reform still remained

in the hands of the state bureaucracy during this period.

Nevertheless it could be argued that competitive pressure on the Congress government

at the state level to implement land reform arose in a manner similar to the post-1978

period. The Congress was aware of the rising political aspirations of the Left front, following

31

the participation of the latter in the 1969-71 United Front ministry in the state. It was

also concerned to restrict the influence of the ultra-left Naxalite party which had incited

a violent conflict in the state in the late 60s. Since the Left did not have any elected

political position in the state during 1974–78, it corresponds to a post-1978 context where

the Congress retained all the political power, i.e., a GP with a zero left share.

Tables 20 and 21 extend the results reported above to including the preceding five year

period 1974–78 (with GP Left share set equal to zero). The estimated coefficients in the

patta regressions with respect to the Left share variables are qualitatively similar, and now

gain in quantitative significance, with a significant negative interaction between competitive

strength and moral hazard. The coefficient of the voter loyalty variable by itself is now also

significant: the Congress enacted more land reform during 1974–78 in places where its

popularity was declining faster (between 1972 and 1977). In addition, there is a significant

pre-election year spike in patta activity. The evidence is therefore consistent with the view

that the 1974–78 Congress administration in charge of implementing land reforms at that

time, was influenced by competition from the rising popularity of the Left.

In the case of the bargadar regression, we continue to find a significant inverted-U with

respect to the Left share variables, a positive election-year spike, and a positive effect of the

voter loyalty variable. The interaction effect continues to be imprecisely estimated, though

its sign is consistently negative.

9 Concluding Comments

In summary, land reform implementation in West Bengal cannot be simply explained by ex-

ogenous differences in redistributive ideology between the Left and the Congress. We found

evidence consistent with the presence of electoral opportunism, and for land reform effort

to increase when electoral contests became less one-sided. We interpret this as implying

that the political will of local governments in West Bengal to implement land reforms was

influenced by electoral institutions, and the evenness of political competition in particular.

It suggests that democratic institutions that promote political competition (e.g., electoral

32

rules that limit the scope for manipulation of electoral outcomes by incumbents) can have

important economic effects on growth and poverty reduction.

Other useful results of our analysis include the fact that electoral success of the Left

in local government elections was driven primarily by factors exogenous to the villages

concerned, pertaining to voter loyalties at the district level, based on economic and political

events at the regional, state and national levels. This permits isolation of effect of a variety

of exogenous events on the extent of land reform implemented. In turn it provides a way to

identify the effect of these land reforms on various aspects of farm productivity and quality

of governance within these villages, a task we are currently undertaking in related research

on the effects of these land reforms.

The main shortcoming of our study pertains to the potential measurement error and

endogeneity bias with respect to controls for land market transactions, migration, fertility,

literacy or caste composition. Given the relatively small scale of the reforms we think the

potential for reverse causation is unlikely, but it cannot be ruled out. There is clearly greater

need to measure these control variables more precisely (though the primary variables of

concern — the land reform and Left share of local governments — are themselves precisely

measured). We are currently carrying out direct household surveys in order to better

measure these and understand patterns of household fragmentation and migration, which

should help to control for these factors better.

References

Arellano M. and S. Bond (1991), ‘Some Tests of Specification for Panel Data: Monte

Carlo Evidence and an Application to Employment Equations’, Review of Economic

Studies 58, 277-297.

Appu P.S. (1996), Land Reforms in India, Delhi: Vikas Publishing House.

Bandyopadhyay D. (1986), ‘Land Reforms in India: An Analysis,” Economic and Political

Weekly, XXI, 25/26: A50-A56.

33

Banerjee A., P. Gertler, and M. Ghatak (2002), ”Empowerment and Efficiency: Tenancy

Reform in West Bengal,” Journal of Political Economy, 110(2), 239-280.

Bardhan P. (1973), “Size, Productivity and Returns to Scale: An Analysis of Farm-Level

Data in Indian Agriculture,” Journal of Political Economy, 81(6), 1370–86.

Basu S.K. and S.K. Bhattacharya (1963), Land Reforms in West Bengal: A Study on

Implementation, Calcutta: Oxford Book Company.

Bell C. (1977), “Alternative Theories of Sharecropping: Some Tests Using Evidence from

Northeast India,” Journal of Development Studies, 13(4), 317–346.

Bergmann T. (1984), Agrarian Reform in India, with Special Reference to Kerala, Kar-

nataka, Andhra Pradesh and West Bengal. New Delhi: Agricole.

Berry A. and Cline W. (1979), Agrarian Structure and Productivity in Developing Coun-

tries, Baltimore: Johns Hopkins University Press.

Bertrand M., E. Duflo and S. Mullainathan (2004), “How Much Should We Trust

Diferences-in-Differences Estimates?” Quarterly Journal of Economics, February 2004,

249–275.

Besley T. and Burgess R. (2000) “Land Reform, Poverty Reduction and Growth: Evidence

from India,” Quarterly Journal of Economics 115, no. 2, 389-430.

Binswanger H., Deininger K., and Feder G. (1993), “Power, Distortions, Revolt and

Reform in Agricultural Land Relations,” in J. Behrman and T.N. Srinivasan (Ed.),

Handbook of Development Economics, vol. III, Amsterdam: Elsevier.

Binswanger H. and M. Rosenzweig (1993), “Wealth, Weather Risk and the Composition

and Profitability of Agricultural Investments,” Economic Journal, 56-78.

Bhattacharya, D. (1999), ‘Politics of Middleness: The Changing Character of the Com-

munist Party of India (Marxist) in Rural Bengal (1977-90),’ in B. Rogaly, B. Harriss-

White and S. Bose (Ed.), Sonar Bangla? Agricultural Growth and Agrarian Change

34

in West Bengal and Bangladesh, Sage Publications., New Delhi and Thousand Oaks,

London.

Case A. (2001), “Election Goals and Income Redistribution: Recent Evidence From

Albania,” European Economic Review, 45, 405-423.

Chatterjee P. (1984), Bengal 1920-47: The Land Question, Calcutta: K.P. Bagchi.

Cooper A. (1988), Sharecropping and Sharecroppers’ Struggles in Bengal, Calcutta: K.P.

Bagchi and Co.

Dixit A. and J. Londregan (1998), “Ideology, Tactics and Efficiency in Redistributive

Politics,” Quarterly Journal of Economics, May 1998, 497–529.

Donald S. and K. Lang (2004), “Inference with Difference in Differences and Other Panel

Data,” mimeo, Department of Economics, Boston University.

Downs A. (1957), An Economic Theory of Democracy, New York: HarperCollins.

Franda M. (1971), Radical Politics in West Bengal, Cambridge, MA: MIT Press.

Grossman G. and Elhanan Helpman, “Electoral Competition and Special Interest Poli-

tics,” Review of Economic Studies, 63, 265-286, 1996.

Honore B. (1992), ‘Trimmed LAD and Least Squares Estimation of Truncated and

Censored Regression Models with Fixed Effects,’ Econometrica, 60, 533-565.

Kohli A. (1987), State of Poverty in India, Cambridge: Cambridge University Press.

Lieten G.K. (1992), Continuity and Change in Rural West Bengal, New Delhi: Sage

Publications.

Lindbeck A. and J. Weibull (1993), “A Model of Political Equilibrium in a Representative

Democracy,” Journal of Public Economics, 51, 195–209.

Lipset S.M. (1960), Political Man. Baltimore: Johns Hopkins University Press.

Nossiter T.J. (1988), Marxist State Governments in India, London: Pinter Publishers.

35

Sengupta S. (1981), ‘West Bengal Land Reforms and the Agrarian Scene,” Economic and

Political Weekly, XVI, 25/26: A26-A75.

Webster N. (1992), Panchayati Raj and the Decentralisation of Development Planning in

West Bengal, Calcutta: K.P. Bagchi and Co.

Wittman D. (1973), ‘Parties as Utility Maximizers,” American Political Science Review,

67, 490–498.

36

Data Appendix: Description of Survey Methods and Data

Our data consists of 88 villages (or more accurately hamlets (mouzas)), spread over 16 out

of 18 districts in West Bengal.21 The villages form a sub-sample of an original stratified

random sample of West Bengal villages selected for a farm survey by the Socio-Economic

Evaluation Branch of the West Bengal Department of Agriculture, for purposes of estimat-

ing agricultural production costs in the state.22 The lowest level of village government is

the gram panchayat (GP), which covers on average 8-12 mouzas. Subsequent to our gath-

ering of farm records, the villages in the sample were surveyed to yield details concerning

composition and activities of local governments, besides patterns of landownership, tenancy,

occupational structure, demographics and literacy. The data includes details of elected GP

representatives for every five year term since 1977 (when GP elections were first held), and

a listing of all households in the village from a list of registered voters for a recent election

year (1998 in most instances) and an earlier year (1978 in most cases, otherwise 1983).

Data concerning the extent of land reforms legally implemented was directly collected

from the local block land records office (BLRO) which contained documents of all land

transfers (pattas) and registration of sharecroppers by village. The date on the documents

revealed the exact timing of the recordings, enabling us to construct a panel data set covering

the period 1971–98. This includes the entire Left Front regime until 1998 spanning four

successive sets of local governments, as well as the preceding five year period of Congress

rule at the state level.

The government’s land records record only the official land transfers, rather than the

landownership distribution at any given point of time. The former is the obvious measure

of the (outcomes of) effort of the government to implement the reforms. But these changes

need to be assessed relative to the existing land distribution (e.g., in constructing percent

21Calcutta and Darjeeling were excluded owing to the paucity of agriculture in those districts: Calcutta is

primarily urban while Darjeeling is a mountainous region dominated by tea plantations. District boundaries

within Dinajpur have changed within the period being studied so we aggregate all the data for Dinajpur

villages. We therefore end up with data for 15 districts.22The sub-sample excludes villages in the original sample for which disaggregated farm-level farm produc-

tion records from the 1970s could not be located at state government offices.

37

of land area that was distributed), and the latter can only be assessed by survey methods.

Efforts to use government land records to construct the landownership distribution within

each village did not succeed, owing to the difficulty of consolidating land titles by households.

We therefore conducted an ‘indirect survey’ whereby three or four village elders provided

details of each household on each voter list concerning land owned, leased or cultivated

(area, irrigation status, mode of acquisition for owned land, barga registration status for

tenants), caste, occupation and literacy status. This provided a complete description of

landownership, occupation and literacy distributions for 1998 and either 1978 or 1983. The

information provided was cross-checked across different elders. This was the only practical

method of constructing the landownership distribution by households within the village

and its change over the past two decades, within the timeframe and budget of the village

surveys.23 The alternative of a direct household survey would have been more expensive,

time consuming and subject to the reluctance of households in remote villages to disclose

their assets to outsiders.24 Our method exploits the fact that landholdings of different

households are well known within the village, and especially to village residents of long

standing. Moreover, our investigators did not perceive any reluctance by elders to disclose

ownership patterns in the village. Moreover, changes in the land distribution provided by

the surveys turned out to correspond closely, when aggregated across villages within the

same district, with district-based data on distribution of operational holdings from the state

Agricultural Censuses.25

23We are currently carrying out a direct household survey in order to ask each household concerning their

landholdings (and changes thereof over the past three decades), but the results will not be available for

research purposes for at least another year or so.24In private communication with us, Debu Bandyopadhyay the Land Reforms Commissioner at the time

when the bulk of the reforms were carried out, expressed his opinion that the indirect survey is likely to be

more reliable than a direct survey for the latter reason.25Details are available in the working paper version of this paper (Bardhan-Mookherjee (2004).

38

TABLE 1: DISTRICT-WISE ALLOCATION OF SAMPLE VILLAGES

DISTRICT NUMBER OF VILLAGES LEFT FRONT

IN SAMPLE PERCENT OF SEATS

IN GP (average 1978-98)

24 Parganas (N) 6 56

24 Parganas (S) 8 54

Bankura 5 87

Birbhum 5 58

Bardhaman 8 84

Cooch-Behar 8 85

Hooghly 6 70

Howrah 4 79

Jalpaiguri 5 74

Malda 2 60

Midnapur 8 78

Murshidabad 6 46

Nadia 5 79

Dinajpur 4 51

Purulia 8 62

WEST BENGAL 88 69

39

TABLE 2: LEFT SHARE IN GP SEATS AND

STATE ASSEMBLY VOTE SHARES

Time Left Front Left front Congress

Block % Seats in % vote in % vote in

in GP Assembly Assembly

(sample) (all WB) (all WB)

1978-83 74 47 23

1983-88 63 53 41

1988-93 71 54 41

1993-98 68 50 34

TABLE 3: CHANGES IN VILLAGE CHARACTERISTICS

1978 AVERAGE 1998 AVERAGE

Number of Households 220 389

Operational Land-household ratio 1.57 acres 0.87 acres

% hhs landless 46.83 51.48

% hhs marginal (0-2.5 acres) 35.27 39.76

% hhs small (2.5-5 acres) 11.41 6.45

% hhs medium (5-12.5 acres) 4.82 2.01

% hhs big (12.5 acres–) 1.66 0.29

% land marginal 27.43 45.75

% land small 28.66 28.17

% land medium 23.57 18.29

% land big 20.34 7.79

% poor hhs illiterate 45.36 32.18

% hhs scheduled castes/tribes 35.33 37.83

% hhs head in nonagri. occupation 39.90 50.26

Farm value added (1974 Rs/acre) 754.04 (in 1981) 1340.16 (in 1996)

Male wage (1974 Rs/hour) 0.65 (in 1981) 0.83 (in 1996)

40

TABLE 4 VESTED LAND: SAMPLE AVERAGES

Vested Land Land above % of

up to 1998 ceiling in 1978 Vested

(BLRO) as % of (survey) as % of land post

operational operational 1978

land in 1998 land in 1978 (BLRO-

(survey) (survey) subsample)

Average village 14.47 6.40 29.43

outside North Bengal

North Bengal village average 19.61 3.67 7.02

All villages average 16.06 5.55 18.38

TABLE 5 PATTA DISTRIBUTION: SAMPLE AVERAGES

(1) (2) (3) (4) (5) (6)

Excluding NB villages 3.76 71.21 90.50 13.77 27.91 54.82

North Bengal village average 9.50 75.22 70.80 18.39 32.60 56.86

All villages average 5.53 73.34 82.98 15.27 29.57 55.60

(1) 1998 Patta land (BLRO) as % of 1998 operational land survey

(2) Post 1978 Patta land as % of total patta land up to 1998 (BLRO)

(3) 1998 Patta cultivable as % of 1998 total patta land (survey)

(4) 1998 Pattadars (BLRO) as % of 1998 HH (survey)

(5) 1998 Pattadars (BLRO) as % of 1998 Landless (survey)

(6) 1998 Pattadars (BLRO) as % of 1978 Landless (survey)

41

TABLE 6: BARGA REGISTRATION: SAMPLE AVERAGES

(1) (2) (3)

Excluding NB villages 7.87 5.21 45.64

North Bengal villages 2.76 3.32 82.61

All villages 6.29 4.60 52.44

(1) 1998 Registered Barga land (BLRO) as % of 1998 operational land

(2) 1998 Registered Bargadars (BLRO) as % of 1998 HH (survey)

(3) 1998 Registered Bargadars (survey) as % of 1998 HH Leasing Land (survey)

TABLE 7: TIME PROFILE OF LAND REFORM

Time # Villages % households % cult. area # Villages % hhs % cult. area

Block distr. pattas rcving pattas distr. rgstrng rgstered rgstered

1978-83 31 9.88 3.37 49 3.26 3.76

1983-88 36 4.68 1.31 32 1.08 0.93

1988-93 35 3.25 0.54 20 0.48 0.24

1993-98 19 1.38 0.18 10 0.13 0.08

TABLE 8: LAND REFORM FREQUENCY 1978-98

Number of Time Blocks Number of Number of Time Blocks Number of

Pattas Distributed Villages Registering Bargadars Villages

4 4 4 3

3 13 3 8

2 19 2 25

1 29 1 26

0 22 0 26

42

43

TABLE 9: VARIATION OF PATTA ACTIVITY

WITH LEFT SHARE OF GP SEATS: TIMEBLOCKS

PATTALAND PATTADAR

Cross Panel Cross Panel

Section Tobit Dist FE Section Tobit Dist FE

% Left 1.15 0.05 1.77 0.36

(1.01) (0.13) (1.23) (0.28)

% Left Sq. -0.92 -0.05 -1.27 -0.35

(0.84) (0.11) (1.03) (0.23)

Total Obs. 85 343 85 344

Censored Obs. 13 225 13 225

Turning Point 63 54 69 52

pattaland: percent cultivable land distributed in pattas

pattadar: percent households receiving pattas

Data aggregated into five year timeblocks representing successive GPs

Tobit estimates, with district fixed effects and time dummies

included in panel, not reported here.

Cross-sections control also for percent land vested in 1978

and population density in 1978

standard errors in parentheses; ***: significant at 1%, ** at 5%, * at 10%

44

TABLE 10: VARIATION OF BARGA ACTIVITY

WITH LEFT SHARE OF GP SEATS: TIMEBLOCKS

BARGALAND BARGADAR

Cross Panel Cross Panel

Section Tobit Dist FE Section Tobit Dist FE

% Left 2.17 0.05 -0.29 0.08

(3.40) (0.16) (0.46) (0.12)

% Left Sq. -1.50 -0.10 0.31 -0.12

(2.83) (0.13) (0.38) (0.10)

Total Obs. 85 346 85 346

Censored Obs. 18 240 18 240

Turning Point 72 27 48 33

bargaland: percent cultivable land registered

bargadar: percent households registered

Data aggregated into five year timeblocks representing successive GPs

Tobit estimates, with district fixed effects and time dummies

included in panel, not reported here.

Cross-sections control also for percent unregistered bargaland per household in 1978

and population density in 1978

standard errors in parentheses; ***: significant at 1%, ** at 5%, * at 10%

45

TABLE 11: PATTA REGRESSIONS, YEARLY DATA 1978-98

DISTRICT/VILLAGE FIXED EFFECTS

PATTALAND PATTADAR

Tobit Tobit TLAD TLAD Tobit Tobit TLAD TLAD

DFE DFE VFE VFE DFE DFE VFE VFE

% Left 3.37e-3 -0.03 0.04 -0.10 0.14 0.11 1.22* 0.38

(8.40e-2) (0.08) (1.08) (0.30) (0.16) (0.16) (0.71) (0.49)

% Left Sq. 4.80e-3 0.02 -0.05 -0.17 -0.12 -0.12 -1.18** -0.48

(6.90e-2) (0.07) (0.79) (0.33) (0.13) (0.13) (0.59) (0.39)

% Election year -9.70e-3 -1.01e-2 -0.08 -0.05 -2.66e-2 2.49e-2 -0.09 -0.11**

Dummy (1.31e-2) (1.32e-2) (0.06) (0.11) (2.56e-2) (2.52e-2) (0.06) (0.05)

% Pre-election year -2.01e-3 -3.61e-3 0.19* 0.17 -1.08e-2 -1.67e-2 0.04 -9.30e-3

Dummy (1.30e-2) (1.30e-3) (0.11) (0.15) (2.52e-2) (2.51e-2) (0.08) (6.88e-2)

% Village Controls* N Y N Y N Y N Y

Total Obs. 1730 1675 1730 1675 1731 1675 1731 1675

Censored Obs. 1556 1501 1556 1501 1556 1501 1556 1501

No. Groups 88 85 88 85 88 85 88 85

* Village controls include land distribution, illiteracy rates, proportion low caste.

Also included: timeblock dummies, and district/village fixed effects.

standard errors in parentheses; ***:significant at 1%, ** at 5%, * at 10%

46

TABLE 12: BARGA REGRESSIONS, YEARLY DATA 1978-98

DISTRICT/VILLAGE FIXED EFFECTS

BARGALAND BARGADAR

Tobit Tobit TLAD TLAD Tobit Tobit TLAD TLAD

DFE DFE VFE VFE DFE DFE VFE VFE

% Left 0.10 0.11 0.38 0.54 0.08 0.08 0.68*** 0.57

(0.10) (0.10) (1.32) (1.89) (0.06) (0.06) (0.23) (0.39)

% Left Sq. -0.14* -0.16* -1.19 -1.22 -0.10** -0.11** -0.63*** -0.55**

(0.08) (0.08) (1.09) (1.49) (0.05) (0.05) (0.20) (0.26)

% Election year 6.30e-3 7.90e-3 0.43** 0.35 2.73e-3 3.88e-3 0.14*** 0.16***

Dummy (1.38e-2) (1.40e-2) (0.18) (0.22) (8.91e-3) (9.01e-3) (0.04) (0.04)

% Pre-election year -1.83e-2 -1.69e-2 -0.02 -0.04 -1.13e-2 -1.03e-2 -7.68e-3 -1.37e-3

Dummy (1.46e-2) (1.48e-2) (0.03) (0.06) (9.38e-3) (9.45e-3) (9.51e-3) (8.30e-2)

% Village Controls* N Y N Y N Y N Y

Total Obs. 1734 1675 1734 1675 1734 1675 1734 1675

Censored Obs. 1576 1517 1576 1517 1576 1517 1576 1517

No. Groups 88 85 88 85 88 85 88 85

* Village controls include land distribution, illiteracy rates, proportion low caste.

Also included: timeblock dummies, and district/village fixed effects.

standard errors in parentheses; ***:significant at 1%, ** at 5%, * at 10%

47

TABLE 13: LEFT SHARE REGRESSIONS

Cross-Section Cross-Section Panel Panel Panel

(OLS) (OLS) (Ar-Bond) (Ar-Bond) (Ar-Bond)

No. obs. (GPs) 55 55 217 (55) 213 (54) 213 (54)

F-st(d.f.) 2.28(8,46) 4.01(9,45) 32.49(6,210) 17.13(14,198) 17.17(15,197)

p-value, 2nd order

ser. corr. diff. res. .61 .24 .26

Assembly Vote Share 1.35*** 0.62**

Difference, District (0.37) (0.25)

Lagged Left Share -0.46** -0.54** -0.48**

in GP (0.20) (0.22) (0.22)

Inflation Rate 2.76*** 2.69*** 3.24***

(0.70) (0.70) 0.72)

Inflation Rate*Lagged -5.53*** -5.35*** -5.83***

GP Left Share (0.89) (0.88) (0.89)

% Cong Seats -0.64*** -0.68*** -0.64***

in Parliament (0.22) (0.24) (0.23)

% Cong Seats Parlmt* 1.32*** 1.37*** 1.38***

Lagged GP LeftShare (0.36) (0.38) (0.37)

Growth Small 0.00 -0.00 -0.01

Fact. Empl., District (0.02) (0.02) 0.02

% HH Landless 0.33 0.35 0.07 0.12

(0.41) (0.37) (0.54) (0.52)

% HH Marginal 0.14 0.24 0.13 0.07

(0.46) (0.41) (0.81) (0.74)

% HH Small 0.07 -0.11 0.08 -0.18

(0.43) (0.39) (0.89) (0.81)

% HH Medium 0.63 0.55 -4.12 -3.33

(0.89) (0.79) (2.58) (2.27)

% Land Small 0.02 0.01 -0.26 -0.18

(0.19) (0.17) (0.42) (0.40)

% Land Big 0.07 -0.04 0.06 0.04

(0.17) (0.16) (0.21) (0.20)

% Poor Illit. -0.19 -0.21** 0.44 0.61

(0.11) (0.10) (0.41) (0.47)

% HH SC/ST 0.36*** 0.28*** -0.01 -0.19

(0.10) (0.09) (0.48) (0.46)

robust standard errors in parentheses; ***: significant at 1%, ** at 5%, * at 10%

48

TABLE 14: AVSD (ASSEMBLY VOTE SHARE

DIFFERENCE) REGRESSIONS

No. ob s. (districts) 75 (15) 75 (15)

F-st (d.f.) 24.26 (7,67) 43.65 (8,66)

p-value, test of 2nd order 0.11 0.10

serial correlation of

differenced residuals

AVSD Lagged 0.36*** 0.35***

(0.12) (0.12)

AVSD second lag 0.21***

(0.06)

Inflation Rate -0.90*** -0.78***

(0.12) (0.14)

% Congress Seats -0.12* -0.16**

in Parliament (0.06) (0.06)

Growth in Small -0.01* -0.01*

Factory Employment (0.01) (0.01)

Growth in Registered 0.08 0.03

Factories (0.07) (0.08)

Growth in Small Factories 0.01 0.01

(0.01) (0.01)

Lagged % Left Front seats -0.01 -0.01

in Assembly (0.00) (0.01)

Arellano-Bond regressions, with district fixed effects

robust standard errors in parentheses; ***: significant at 1%, ** at 5%, * at 10%

49

TABLE 15: IV PATTA REGRESSIONS, YEARLY DATA, 1978-98

PATTALAND PATTADAR

Tobit TLAD Tobit TLAD

DFE VFE DFE VFE

% Pred. LF in GP 2.97e-4 0.07 0.23 0.76

(1.13e-1) (1.06) (0.22) (1.09)

% Pred. LF Sq. 3.42e-2 0.41 -0.10 -0.23

(9.11e-2) (0.71) (0.18) (0.68)

Election Year -1.22e-2 -0.08 -2.88e-2 -0.13**

Dummy (1.35e-2) (0.10) (2.58e-2) (0.06)

Pre-election Year -5.52-3 0.15 -1.99e-2 -0.02

Dummy (1.34e-2) (0.16) (2.56e-2) (0.07)

Turning Points n.a. n.a. 108 168

Total Obs. 1660 1660 1660 1660

Censored Obs. 1489 1489 1489 1489

No. Groups 84 84 84 84

Village characteristics and timeblock dummies included, not reported here

***: significant at 1%;**:significant at 5%;* at 10%

50

TABLE 16: IV BARGA REGRESSIONS, YEARLY DATA, 1978-98

BARGALAND BARGADAR

Tobit TLAD Tobit TLAD

DFE VFE DFE VFE

% Pred. LF in GP 0.23 4.18 0.16* 2.85***

(0.14) (4.90) (0.09) (0.97)

% Pred. LF Sq. -0.17 -2.41 -0.12 -1.72***

(0.12) (3.75) (0.07) (0.65)

Election Year 8.13e-3 0.31* 4.00e-3 0.14***

Dummy (1.39e-2) (0.16) (9.00e-3) (0.05)

Pre-election Year -1.58-2 -0.08 -9.59e-3 -7.10e-4

Dummy (1.46e-2) (0.06) (9.39e-3) (1.45e-2)

Total Obs. 1660 1660 1660 1660

Censored Obs. 1502 1502 1502 1502

No. Groups 84 84 84 84

Turning Points 68 87 67 83

Village characteristics and timeblock dummies included, not reported here

***: significant at 1%;**:significant at 5%;* at 10%

51

TABLE 17: TLAD REGRESSIONS FOR PATTA ACTIVITY

INCL. AVSD AND INTERACTIONS, YEARLY DATA 1978-98+

PATTALAND PATTADAR

Non-IV IV Non-IV IV

% LF in GP 1.30 1.86 2.36 3.42

se (2.05) (1.27) (1.36)* (1.90)

bse (2.05) (3.19) (1.75) (1.84)*

% LF sq. -1.98 -1.67 -2.66 -2.99

se (1.69) (0.96)* (1.27)** (1.51)**

bse (1.72) (2.32) (1.52) (1.48)**

AVSD -0.29 2.96 0.18 3.11

se (5.69) (4.70) (1.36) (2.83)

bse (6.11) (5.79) (3.91) (3.40)

AVSD*LF -11.60 -19.74 -13.67 -21.38

se (18.59) (12.01) (8.20)* (9.74)**

bse (18.88) (17.95) (12.62) (10.87)**

AVSD*LF sq. 13.52 17.32 14.76 18.62

se (13.81) (7.73)** (7.80)* (7.50)**

bse (14.29) (13.00) (10.09) (8.29)**

Election year Dummy -0.05 -0.05 -0.10 -0.11

se (0.10) (0.08) (0.06)* (0.06)*

bse (0.12) (0.10) (0.06) (0.06)*

Pre-election yr. Dummy 0.18 0.16 0.003 -0.03

se (0.15) (0.14) (0.06) (0.06)

bse (0.19) (0.20) (0.07) (0.07)

Total Obs. 1675 1660 1675 1660

Censored Obs. 1501 1489 1501 1489

No. Groups 85 84 85 84

Turn Pt: % LF 33 56 45 57

Turn Pt: 43 57 46 57

AVSD*% LF

+ For coefficients of village characteristics in IV regressions see Table 19.

se: Standard errors, bse: Block-bootstrapped standard errors

Village characteristics and timeblock dummies included, not reported here

***: significant at 1%;**:significant at 5%;* at 10%

52

TABLE 18: TLAD REGRESSIONS FOR BARGA ACTIVITY

INCL. AVSD AND INTERACTIONS, YEARLY DATA 1978-98+

BARGALAND BARGADAR

Non-IV IV Non-IV IV

% LF in GP 4.65 10.85 0.66 4.30

se (3.06) (10.51) (0.46) (2.02)**

bse (3.30) (7.94) (0.95) (1.91)**

% LF sq. -4.82 -6.90 -0.79 -2.68

se (2.77)* (7.31) (0.38)** (1.35)**

bse (2.88)* (5.51) (0.77) (1.37)**

AVSD 4.54 14.80 -0.82 4.31

se (5.34) (15.94) (0.89) (3.23)

bse (6.82) (11.71) (1.82) (3.10)

AVSD*LF -32.46 -51.72 0.36 -12.44

se (23.35) (46.14) (3.20) (9.15)

bse (26.23) (38.07) (6.37) (9.33)

AVSD*LF sq. 27.56 35.41 0.88 8.04

se (18.96) (31.57) (2.52) (6.30)

bse (20.19) (26.78) (4.95) (6.67)

Election year Dummy 0.30 0.27 0.16 0.14

se (0.19) (0.14)* (0.04)*** (0.05)***

bse (0.17)* (0.17) (0.06)*** (0.06)**

Pre-election yr. Dummy -0.13 -0.16 -0.0013 -0.003

se (0.10) (0.09)* (0.01) (0.02)

bse (0.10) (0.11) (0.02) (0.02)

Total Obs. 1675 1660 1675 1660

Censored Obs. 1517 1502 1517 1502

No. Groups 85 84 85 84

Turn Pt: % LF 48 79 42 80

Turn Pt: 59 73 n.a. 77

AVSD*% LF

+ For coefficients of village characteristics in IV regressions see Table 19.

se: Standard errors, bse: Block-bootstrapped standard errors

Village characteristics and timeblock dummies included, not reported here

***: significant at 1%;**:significant at 5%;* at 10%

53

TABLE 19: VILLAGE CHARACTERISTICS, IV TLAD REGRESSIONS

INCL. AVSD AND INTERACTIONS, YEARLY DATA 1978-98+

PATTALAND PATTADAR BARGALAND BARGADAR

% HH Landless -2.86 -0.08 0.72 -2.94

se (4.79) (1.01) (3.01) (1.99)

bse (5.53) (2.69) (4.35) (1.69)*

% HH Marginal -1.64 2.86 -2.18 -2.19

se (3.45) (1.10)*** (2.49) (1.91)

bse (5.50) (3.16) (4.20) (1.60)

% HH Small -5.88 0.26 3.58 -0.50

se (4.16) (1.17) (5.85) (2.37)

bse (6.52) (2.69) (4.70) (1.24)

% HH Medium -13.31 -8.75 -1.99 -1.52

se (8.62) (2.74)*** (2.54) (1.38)

bse (12.80) (5.07)* (4.63) (1.55)

% Land Small -2.19 -2.72 0.83 -0.20

se (1.15)* (0.90)*** (1.10) (0.35)

bse (3.08) (1.26)** (1.47) (0.38)

% Land Big 0.33 0.29 -0.05 -0.12

se (0.30) (0.32) (0.19) (0.15)

bse (1.31) (0.52) (0.57) (0.18)

% Poor Illiterate -2.02 -1.03 -0.05 0.72

se (1.25) (0.64) (0.75) (0.21)***

bse (1.74) (0.67) (1.03) (0.33)**

% HH SC/ST -7.91 -0.62 -4.91 -0.66

se (2.44)*** (1.08) (2.17)** (1.70)

bse (5.12) (1.37) (3.86) (1.07)

+ Coefficients of village characteristics in IV regressions in Tables 17 and 18.

***: significant at 1%;**:significant at 5%;* at 10%

54

TABLE 20: TLAD REGRESSIONS FOR PATTA ACTIVITY

INCL. INTERACTION EFFECTS, YEARLY DATA 1974–98

PATTALAND PATTADAR

Obs. % LF Inst.% LF Obs. % LF Inst.% LF

T-LAD VFE T-LAD VFE T-LAD VFE T-LAD VFE

Total Obs. 2100 2085 2100 2085

Censored Obs. 1885 1873 1885 1873

% LF in GP 1.84* 3.86*** 2.60** 4.44***

(1.06) (1.41) (1.07) (1.65)

% LF sq. -1.93** -2.98*** -2.68** -3.90***

(0.96) (1.06) (1.04) (1.36)

AVSD 1.79*** 1.86*** 1.99*** 2.04***

(0.45) (0.60) (0.67) (0.73)

AVSD*LF -8.65*** -10.78*** -12.63*** -16.00***

(3.33) (3.37) (4.63) (3.63)

AVSD*LF sq. 9.76** 11.47*** 12.54** 16.29***

(4.39) (3.43) (5.03) (3.72)

Election year -0.001 -0.003 -0.10* -0.09

Dummy (0.04) (0.05) (0.06) (0.06)

Pre-election 0.24* 0.24* 0.11* 0.10*

year Dummy (0.13) (0.12) (0.06) (0.06)

Turn Pt: % LF 48 65 48 57

Turn Pt: 44 47 50 49

AVSD*% LF

Village characteristics and timeblock dummies included, not reported here

***: significant at 1%;**:significant at 5%;* at 10%

55

TABLE 21: TLAD REGRESSIONS FOR BARGA ACTIVITY

INCL. INTERACTION EFFECTS, YEARLY DATA 1974–98

BARGALAND BARGADAR

Obs. % LF Inst.% LF Obs. % LF Inst. % LF

T-LAD VFE T-LAD VFE T-LAD VFE T-LAD VFE

Total Obs. 2100 2085 2100 2085

Censored Obs. 1897 1882 1897 1882

% LF in GP 3.65** 5.62 1.03 1.61***

(1.73) (3.60) (0.64) (0.60)

% LF sq. -3.91** -4.04 -1.10* -1.26***

(1.63) (2.53) (0.58) (0.49)

AVSD 0.94 -1.47* 0.42** 0.64***

(0.64) (0.82) (0.21) (0.22)

AVSD*LF -15.35 -10.41 -2.29 -1.07

(8.43) (7.17) (2.04) (1.57)

AVSD*LF sq. 14.88 8.17 2.51 0.46

(8.14) (7.22) (2.32) (1.64)

Election year 0.48* 0.44 0.17*** 0.17***

Dummy (0.29) (0.32) (0.05) (0.04)

Pre-election -0.15 * -0.19** -0.03 -0.04

year Dummy (0.09) (0.08) (0.02) (0.02)

Village characteristics included, not reported here

***: significant at 1%;**:significant at 5%;* at 10%

LEFT’S WINPROBABILITY

LANDREFORM

RIGHT PARTY’S POLICY

LEFT PARTY’S POLICY

Figure 1: Case 1: Political Moral Hazard Effect Dominates

LEFT’S POLICY

RIGHT’S POLICY

LEFT’S WINPROBABILITY

LANDREFORM

Figure 2: Case 2: Ideology Dominates

56