Does Sharecropping Affect Long-term Investment? Evidence from West Bengal's Tenancy Reforms

Land Reform in India: Conceptual Basis and Implementation Experience

While the negative long-term impacts of unequal asset distribution have long been recognized (24; 28), attention has only recently focused on the distribution of resources and political power as being key determinants of institutional development that have far-reaching impacts on socio-economic and human development (1; 50). Land and agricultural institutions including the inequality of land ownership and whether cultivators had full ownership rights to their land had far-reaching impacts on agricultural productivity and the supply of public goods in India (9; 38), Central America (49), Brazil (48), as well as the United States (43).

If highly unequal asset distributions are maintained by coercion or other non-economic means, broader political and economic changes often make it difficult to maintain these distributions, thus creating a need for reform (6; 24; 55; 59). How reform is implemented also has far-reaching implications for productive performance: While it often increases incentives for land-related investment and land use productivity (45), failure to quickly implement measures can lead to prolonged phases of insecurity and strife, with negative impacts on land-related investment, productivity, and broader economic development in many cases (19).

Land reform was high on India's immediate post-independence policy agenda after 1948, resulting in a swift and successful abolition of rent-collecting intermediaries (zamindars). Two other measures introduced by the 1955 Land Reform Act were: (1) the ability to expropriate any land held by a household above a given ceiling and vest it with the state for subsequent redistribution to the poor; and (2) tenancy laws that provided sitting tenants with permanent and registered use rights. Both policies were controversial, difficult to implement,¹ and had undesired adverse side effects, as landlords not only evicted tenants to prevent them from becoming eligible for such reforms (4), but also failed to enter into new lease agreements with them.² Overall reform impacts have been found to be ambiguous, depending on state level implementation (18; 27; 35). Enforcement was a key reason for the ambiguity: although the law provided tenants with permanent and inheritable use rights, landlords' ability to seize land for “personal cultivation” created a loophole that was routinely used as a pretext for evicting tenants. To enjoy their legal rights, tenants had to formally register them; but intimidation by landlords—who often also controlled the local government authorities where registration had to take place—as well as a lack of knowledge prevented many from doing so (22).

In West Bengal, the 1978 election of the Left Front created the right conditions for more effective land reform implementation. While outright transfer of property rights to tenants was not politically feasible, an aggressive program to register tenants was launched in 1978 (44; 10). This effort, driven by local-government, was relatively equitable and benefited some 4 million households (13). By resolving earlier uncertainty, the program increased investment incentives, bringing about rapid productivity growth (53). Although data do not always allow reform effects to be isolated from impacts of other interventions, studies find positive impacts, with estimated gains of between 51% and 63% (10). Others point towards large general equilibrium effects, but differences in the magnitude of impacts between tenancy reform and transfers of vested land, i.e. land expropriated from land owners who held land in excess of the ceiling (14).

Conceptually, tenancy reform comprises three main elements (7). First, registered tenants (bargadars) are protected from eviction as long as they pay rent; in fact, share tenancies (bargas) are made heritable. Second, fixed-rent contracts are outlawed and ceilings of 25% or 50%, depending on whether the landlord supplies inputs, are imposed on the share of rent that a tenant can be charged. Finally, the transfer of tenanted land (barga) by the tenant to third parties via subleasing is not allowed. A detailed study of the productivity impacts of such measures is not only relevant for West Bengal, but also allows us to obtain an unbiased estimate of the productivity impact of sharecropping.

Conceptual Basis and Empirical Estimates of the Size of the “Marshallian Inefficiency”

In a world of perfect information, complete markets, and zero transaction costs, the distribution of land ownership would affect welfare but would not matter for efficiency, as market transactions allow everybody to attain an optimum farm size (32). Imperfections in labor as well as credit and insurance markets change this, and together with contracting parties' attributes and transaction costs related to contract enforcement, will affect productivity outcomes from land rental (20). Most importantly, although cash rental would maximize productivity under ideal conditions, share contracts may arise if there is non-observable effort or limited liability, and tenants are risk-averse (51; 58). Although characterized by lower productivity than fixed rent, share tenancy may then be chosen as a second-best option in specific environments (16; 34).³

Many empirical studies have aimed to quantify the “Marshallian inefficiency” of sharecropping, and have obtained varying results that can be partly explained by differences in methodology (17; 57; 52; 25; 40). Even studies controlling for household fixed effects come to widely divergent conclusions, with estimates ranging from a productivity loss of 16% (57), to impacts that, while statistically significant, appear to be of limited economic relevance (40).⁴ Obtaining credible estimates requires that, in addition to controlling for unobserved household characteristics, two challenges be addressed. First, contractual arrangements vary with tenants' capital endowments (41), ability (42), farming experience (23; 29), and the relevance of permanent structures or the need to preserve soil fertility (30; 54). Thus, contract choice will be endogenous in many settings. Moreover, beyond its short-term impact, sharecropping may have longer-term effects. Specifically, in a long-term relationship, the ability to develop one's reputation will, as in any repeated game, decrease the level of inefficiency observed in any given period. On the other hand, the fact that neither party can fully appropriate the surplus from long-term investment can reduce investment incentives by landlords and tenants compared to a situation where one party owns the land. Even if, in a given setting, it is a second-best outcome, a reliable quantitative estimate of the productivity impact of sharecropping can provide an upper bound for the benefits from, say, eliminating the conditions (e.g. credit market imperfections) that give rise to such arrangements.

Analytical Strategy and Hypotheses

In West Bengal, restrictions on contract choice imposed in 1978 made it undesirable to enter into new lease contracts; this creates a situation where decisions about share tenancy arrangements and their modalities were made long ago, and can thus plausibly be considered exogenous. There are three reasons for this: First, landlords are unlikely to enter into new (formal) lease agreements, because any new tenant would become eligible for registration and thus receive long-term tenancy rights, a notion supported by a precipitous decline in rentals. Second, short-term land transfers by (protected) tenants are precluded by prohibitions on sub-leasing that threaten those leasing out with the loss of their use rights. Finally, a 25% or 50% sharing rule, depending on who provides inputs, are the only contractual options allowed by law.

The fact that contractual arrangements respond to exogenous legal constraints rather than being chosen by contracting parties reduces concerns about the endogeneity of contract choice that affect most studies in this area. This makes our setting well-suited to provide robust estimates of the contemporaneous and long-term productivity effects of sharecropping. This is policy-relevant because in the absence of other binding constraints, the productivity losses from share tenancy provide an upper bound for the gains possibly to be realized from eliminating the conditions (legal restrictions or credit market imperfections) that cause such arrangements.

To frame the empirical analysis, we consider how West Bengal's policies may affect productivity and investment. Rent ceilings and awarding permanent use rights to their plots will increase tenants' wealth and bargaining power vis-à-vis landlords, which is consistent with the observation that tenant registration led to a drop in the proportion of contracts with a landlord by half, and a concomitant increase of those with a 75/25% sharing rule (10). The implied increase in investment incentives is likely to outweigh any potential negative impacts arising from the reform-induced elimination of eviction threats as a device to increase tenants' incentives for effort, which may reduce productivity for short-term contracts (11). Also, the long-term nature of contracts will transform tenant-landlord bargaining into a repeated game, possibly increasing the incentive to apply optimum levels of inputs and efficiency of production in any given period. This implies that our results are likely to constitute a conservative estimate of the contemporaneous productivity-impact of share tenancy.

Regarding longer-term investment incentives, two factors have to be considered. First, reform-induced wealth transfer will increase tenants' ability to undertake indivisible investments. At the same time, as output has to be shared, neither party to the contract can reap the full marginal benefits of such investment. Thus, unless parties reach an implicit and enforceable agreement on how to share benefits, investment levels are likely to be below what would be expected under full ownership. The long-term nature of contractual relationships in West Bengal implies that any effects will be lower than expected in a setting where sharecropping contracts are renewed annually, and contractual parameters are less constrained.

Estimation Strategy

To quantify the effect of sharecropping on agricultural production revenues and the probability of making long-term land-attached investments, we follow 17 and 57, and use plot-level data on production and investment from our sample of owner-cum tenants to estimate:

(1)

where Y_ij is the outcome of interest, either the value of gross or net revenue, excluding family labor and level of input use per acre, or an indicator of investment having been made during the past on plot j by cultivator i. Moreover, α_i captures the unobserved determinants of productivity by i including farming skills, access to technology and credit, etc; R_ij is an indicator variable for sharecropped plots; and X_ij is a vector of observable plot characteristics including size, soil type or quality, access to irrigation and drainage, and distance from the operator's homestead.⁵ Empirically, identification and truncation issues need to be considered.

For the first set of concerns, as α_i will be correlated with R_ij (or E(α_i|R_ij=1)≠0), the Ordinary Least Squares (OLS) estimation of β will be biased. To deal with this, we follow other studies using plot-level data (40; 57), and limit our sample to owner-cum-tenants so that β can be identified from within household variation.⁶

A second source of bias is that E(ε_ij|R_ij=1)≠0; in fact, sharecropped plots are often assumed to be of lower quality than owned ones. Two factors alleviate concerns that such bias may affect our estimates. First, we control for a wide range of observable plot attributes. Second, many unobserved soil quality attributes such as texture, capillarity, drainage, and salinity respond to farmers' management decisions over longer periods of time and can thus be considered an investment. As the acquisition of new barga plots came to a virtual standstill after 1978, differences in these attributes can at least partly be attributed to tenants' actions since receiving the plot so that differences would be an indicator of tenure-induced under-investment. If unobserved differences existed in soil quality, they would likely lead us to overestimate the magnitude of the “Marshallian inefficiency.”

Although such a bias is unlikely in our setting, it is useful to bound the potential magnitude of such a bias under rather extreme assumptions. To do so, we use the methodology developed by 3 for categorical variables.⁷ The lower bounds are estimated based on the assumption of equality of selection on both observables and unobservables. Specifically, values are obtained by estimating a bivariate probit model for investment decision and tenancy status of plot j by imposing ρ=Cov(X'β, X'γ)/Var(X'γ), where ρ is the correlation between the error components in the investment decision equation I=1(X'γ+α*T+ε>0) and the share tenancy equation (T=1(X'β+u>0)).

Moreover, our data will include a large amount of zeros for investment and some types of inputs. We thus adopt the semi-parametric trimmed LAD approach (37) to estimate a fixed-effect Tobit to complement the linear probability model used to ascertain the probability of any investment being made. If investment incentives are systematically lower on sharecropped land than on owned land, the total productivity effect of barga tenure will be the sum of “Marshallian inefficiency” and investment effects. For observable investments such as irrigation that enter into the production function, multiplying relevant coefficients is one way to estimate total productivity impacts.

Household-level evidence from listing data

Our data is from a 2008/9 World Bank/FAO survey conducted in 200 randomly-selected villages from 10 districts in West Bengal. Villages were selected randomly with probability of selection being proportional to the number of beneficiaries from the 1978 land reforms, based on official lists obtained from the State Institute of Panchayats & Rural Development (21). The data is thus representative of the universe of West Bengal's land reform beneficiaries. A complete listing of all village residents provided, for the approximately 96,000 households in our 200 sample villages, basic information on current and past household structure, key asset endowments, and the extent and nature to which households were affected by land reform. This was followed by an in-depth follow-up survey of some 1,800 owner-cum-tenants to compare productivity, intensity of input use, and investment between owned plots and those for which inheritable barga rights were received during the 1978 reforms.

The village-wide sample highlights that, after the reforms, few if any new sharecropping contracts were concluded. The majority (82%) of share tenants in our sample thus either held their land since 1978, or inherited tenancy rights to a given plot of land (barga) from family members who benefited from land reform. In the remaining 18% of cases, land already encumbered with a tenant had been sold to a third party without interrupting pre-existing tenancy relationships. In line with descriptive studies (36), the data include 484 cases where residual claims by the other party were bought out by one of the parties, either by the tenant (289 cases) or the landlord (195 cases).

Table 1 reports key 1978 household characteristics, based on recall, for the overall sample and landless, pure owners, owner-cum-tenants, and land reform beneficiaries on tenanted land (bargadars) or ceiling surplus land (pattadars). The table implies that, as is well documented (12), land reform targeted the less well-off. For example, the share of illiterate heads of household was 70% overall, but 85% for land reform beneficiaries, 76% for the landless, and 67% for owner-cum-tenants. This is mirrored by the household head's formal education, which ranges from 4 years for owners to less than 2 years for reform beneficiaries and landless who are also more likely to live in houses with thatch or plastic roofs and bamboo or mud walls. Reform beneficiaries also had no or little land of their own and were more likely to come from Scheduled Castes or Scheduled Tribes.

Current socio-economic characteristics in table 2 imply that land reform helped to somewhat—but by no means drastically—improve beneficiaries' socio-economic position. Some 19% of household heads among pure bargadars, or 16% among owner-cum-tenants, had their main occupation listed as non-farm jobs (including off-farm wage and off-farm self-employment), compared to 27% of pure owners and 53% among the landless. Among beneficiaries, owner-cum-tenants rely more on agriculture than pure bargadars, most likely due to a higher land endowment compared to pure tenants (2.35 vs. 1.41 acres). Beneficiaries from tenancy reform had lower incomes than non-beneficiaries (4,352 Rs./capita for pure tenants, and 4,665 for owner-cum-tenants vs. Rs. 5,914 for pure owners, and Rs. 5,469 for landless). As one would expect where rental restrictions are binding, the restrictions imposed by land reform legislation impose imbalances on the land rental market. Compared to almost two-thirds (63%) of sample households who want to rent in (78% of pure bargadars and pattadars), only 1% indicate having rented out land, and only 3% are interested in leasing out.⁸ Beyond potential Marshallian inefficiency, reform-induced restrictions on land rental may thus reduce the scope for rural productivity growth (26). The potential inefficiency of tenancy is also supported by listing data. Gross and net crop revenues are higher for land owners (21,628 and 9,112 Rs./ac) than for owner-cum-tenants (18,167 and 8,178 Rs./ac). And gross and net crop renenue are lowest for pure tenants (13,217 and 5,556 Rs./ac, respectively).⁹

Transition matrices from the listing data for the total sample, and for households formed before and after 1978, respectively, are reported in table A1. We note a slight decline in the number of landless (54% to 49% of the sample) and the share of owners (from 37.6% to 34.7%), while there is a slight increase in bargadars (2.6% to 3.4%) and owners-cum-tenants (2.3% to 3.4%). For the 15,399 households (16.5% of the sample) that had existed since 1978, the numbers of bargadars, pure owners, owner-cum-tenants, and pattadars all increased (by 241, 322,and 502, and 1,205, respectively), while the landless decreased from 49% to 34%, suggesting some “upward” movement. For dynasties that have since split apart (84% of the sample, or 80,527 households), the number of landless, owner-cum-tenants, and bargadars increased slightly (by 2,665, 1,013, and 777), pattadars more than doubled (2,584 to 6,817), and the number of pure owners dropped correspondingly. Compared to other post-reform settings such as Taiwan or Korea during a comparable period, movement in the 30 years since land reform seems limited.

Plot-Level Data for the Owner-Cum-Tenant Sample

Detailed information on plot characteristics is available for the sample of owner-cum-tenants as reported in table 3. Plots measure 0.4 acres on average (0.36 vs. 0.46 acres for owned and barga, respectively) with barga plots being about 170 m more distant from the cultivator's home. Permanent plot characteristics such as soil type, color, or condition are nearly indistinguishable between owned and barga plots. Although magnitudes are small, barga plots are more likely affected by salinity and drainage problems caused by inadequate management over a long time, possibly a result of investment disincentives: 46.6% (50.6%) of own compared to 45.2% (52%) for barga plots have no (moderate) salinity problems. Similarly, 35.4% (14.1%) of own plots are “easy (difficult) to drain” compared to 34% (15.2%) for barga plots.

Productivity per acre on tenanted plots is significantly lower than on owned plots, by some 24% for gross and net revenue, excluding family labor (Rs 16,693 vs. Rs. 22,062 and Rs 11,051 vs. Rs 14,565, respectively). Interestingly, and consistent with predictions, this difference in net revenues is not due to higher levels of input use on barga plots; on the contrary, the intensity of fertilizer, pesticide, seed, draught power, and family labor use are significantly lower on barga plots, as one would expect.

Investment data suggest that land-attached investment is much lower on barga than on owned plots. Although access to irrigation is a rough measure, we find a 10-point difference in access to private irrigation, largely bore wells, which exist on 54% of owned plots and 44% of barga plots. By contrast, there is no difference between owned and barga plots in access to public irrigation.¹⁰ Flows of less observable investments to maintain or improve land quality point in the same direction: Over the last 8 years, such investment was undertaken on 39% of owned vs. 12% of barga plots. Amounts of capital and family labor spent on such investments in 2007 are 3 and 4 times larger than on the former (Rs. 203 vs. 51 and 6.8 vs. 1.8 days, respectively). Econometric analysis can help to more clearly assess the magnitude of tenancy-induced effects.

Cross-sectional Evidence From Listing Data

As descriptive data point towards differences in observables between owner-cum-tenants and others, checking for potential differences in unobservable attributes that may set this group aside and possibly compromise the external validity of results from our plot-level analysis is needed. To accomplish this and at the same time obtain a proxy of overall effects of share tenancy on output and investment, we use listing data to run cross-sectional regressions for all 40,000 agricultural cultivators. Two specifications are used, one with a zero-one indicator each for pure tenants and owner-cum-tenants, and one with the share of tenanted land in a household's total cropped area as independent variables on the right-hand side.

Results for output from production in the current period (columns 1–4 of table 4) suggest that, irrespective of whether gross or net revenue is used as a dependent variable, tenancy has a statistically significant and economically meaningful impact on productivity. The point estimate for the magnitude of the “Marshallian inefficiency” for pure tenants is close to 20% in both cases. At the same time, with estimates of 11% and 14%, the coefficient for owner-cum-tenants is significantly smaller (columns 1 and 3); in fact, we can reject the hypothesis of the two coefficients being equal at the 5% significance level. This implies that pure tenants are less efficient than owner-cum-tenants, possibly because owner-cum-tenants' owned plots will not be subject to “Marshallian inefficiency.” Including the share of land tenanted directly on the right-hand side supports this; estimates suggest that gross and net revenue for pure tenants will be lower by 20% or 23% than that of full owners; this is similar to what theory would lead us to expect. Although the level of investment in the sample is low overall, equivalent regressions on a zero/one indicator of whether any investment was undertaken since the household was established support the notion of significantly lower investment on tenanted plots compared to owned plots. The coefficient for pure tenants, though small quantitatively, is marginally significant and negative. There is no evidence of investment being significantly lower for owner-cum-tenants; to the contrary, the point estimate is positive, possibly because this group faces few obstacles to investing in owned plots. A similar result emerges if we use the share of tenanted land. All of this makes it unlikely that unobserved differences lead to systematic differences between owner-cum-tenants' and pure tenants' behavior. The sample should thus allow us to reliably estimate the impact of sharecropping as a conservative estimate of the impacts of such an arrangement in environments where contract choice is less constrained by regulations.

Within-Household Estimates of Tenancy Effects on Productivity and Input Use

Table 5 reports results from the household fixed-effect regressions for gross or net revenue from crop production on a given plot. Including household fixed effects limits the set of explanatory variables to plot size and distance, irrigation status, and soil quality indicators, in addition to the barga dummy. The coefficient on the latter, which, as the dependent variable is in logs, can be interpreted as an elasticity, is significant throughout, suggesting that net and gross revenue on barga plots are 19 and 20 points lower than on owner-cultivated plots in the same household. The coefficient on distance to the homestead is negative as expected, suggesting that gross or net revenue for plots owned by the same household that are located 1 km further away is lower by 7% and 9%, respectively. Plot size is estimated to have no appreciable impact on gross revenue but does slightly increase net revenue possibly due to economies of scope that decrease the cost of cultivating larger plots (33).

Results are robust to different specifications. Adding plot and soil characteristics (columns 2 and 4) leaves our main result virtually unaffected. Limited within-household variation in sharing rules, together with the fact that most households obtain land from only one landlord, makes it difficult to assess whether and how much landlord monitoring or provision of inputs might help attenuate “Marshallian inefficiency.”¹¹ Overall, the output effect obtained here is comparable to the 16% estimated by 57, and to the 22% obtained for West Bengal by Bardhan (2009).¹²

Although it is much larger than the result by 40 for Pakistan, the result is comparable to their estimate of 18% for plots not subject to landlord supervision. As close supervision entails costs the magnitude of which is likely to increase with economic development, our results, which will constitute a lower bound estimate, suggest that disincentive effects from contractual arrangements in land markets may warrant attention from policy-makers.

Results from fixed effect Tobit estimates for the level of input use in table 6 are consistent with earlier results and, in addition, suggest an interesting variation across inputs depending on how much their application can be observed. For all inputs except seeds which are easily observed, amounts applied per acre on tenanted plots are significantly below those on owned plots, as predicted by theory. Comparing coefficients with mean levels of application suggests differences in magnitude, which ranges from 11% for family labor, to 20% for pesticide and 28% for fertilizer. With the exception of seeds and casual labor, inputs are applied less intensively on larger plots, possibly due to the cost savings associated with their cultivation. Fertilizer and pesticides are also used less intensively on plots further away from the home.

Investment Effects and Implications for Land Values

To explore investment impacts, we use zero/one indicators for whether investment in land improvement was undertaken during the last 8 years, and whether family labor was used for such investment. We also use amounts of money and labor spent on such investment, and indicators for whether or not a plot had access to public or private irrigation. Results from a linear probability model (columns 1 and 2 of table 7) suggest that investment incentives by owner-cum-cultivators are significantly lower on tenanted plots than on owned plots, contrary to the objective of land reform, which was to enhance investment.¹³ The average barga plot is 26% and 22% less likely than a plot owned by the same household to have received investment to maintain or improve land quality over the last 8 years, and to have used family labor for such investment, respectively, which indicate considerable levels of tenure insecurity or inability to enforce contracts. As one would expect, estimates of the effect of share tenancy on the likelihood of land investment under the assumption of bias due to the presence of unobserved differences in soil fertility beyond those controlled for in our regression using the method by 3 are smaller than those in the fixed-effect model. Estimated effects on the propensity to have undertaken any land improvement during the previous 8 years, and to have used family labor in 2008, are 10% and 7%, respectively, in both cases still significant at the 1% level, thus supporting our main result.¹⁴

Estimates from the fixed-effect Tobit model for actual cash and labor days spent on land improvements during the last 8 years or 12 months, as reported in the last two columns of table 7, are also negative and significant at the 1% level. Coefficients suggest that sharecropped plots receive 21 days of family labor and Rs. 959 (or Rs. 120/year) less in terms of investment than do plots owned by the same household. Compared to soil fertility improvements, investment in private irrigation is more capital-intensive and visible, and may generate external effects due to indivisibilities, which could also increase landlords' incentives to undertake such investment.¹⁵ Column 5 of table 7 shows that, although the coefficient on access to private irrigation is smaller than that on land investment, it is negative and highly significant, suggesting that barga plots are 7% less likely to have private irrigation equipment than owned plots, in addition to private irrigation being less likely on plots further from the homestead. As we do not expect plots' tenure status to affect the placement of public irrigation, a similar regression for public irrigation can serve as a placebo. Indeed, doing so yields an insignificant coefficient (column 6 of table 7), thus increasing confidence in the data and allaying fears that other unobserved factors may drive our results.

Taken together, these estimates indicate significant investment disincentives on tenanted plots, both in terms of the size of the estimated impact and the nature of investments considered. These are larger than what had been found in other studies exploring commitment problems and their impact on under-provision of non-contractible investment (39).

Lack of information on productivity impacts of land-attached investment makes it difficult to translate such tenure-induced under-investment into productivity impacts. However, a rough lower-bound estimate can be obtained if we neglect any productivity-impacts of soil improvements, that is, we assume that barga tenure will only affect private irrigation investment. In this case, the estimated impact of tenure-induced underinvestment would reduce gross or net revenue by 3.4–4.4%.¹⁶ Adding this to the estimate of “Marshallian inefficiency” obtained earlier would imply a total efficiency loss from barga tenure of close to 25%. This is not only a large effect, it also suggests that, if other factors such as credit access are not binding, ways to permit a buyout of the other party's residual interest (46) could help increase productivity without requiring access to large amounts of public finance, as predicted productivity increases would be large enough to pay for the associated cost.

Since our survey also asked respondents about the amount for which they would be able to sell a given plot, similar within-household regressions allow us to assess the impact of tenure on hypothetical sales prices, and thus obtain an upper bound on tenure-induced productivity effects. Results in table 8 suggest that prices for plots tenanted by the same household are some 42% below those that are owned.