Indirect Effects of Land Reform - Regression Results

5 Indirect Effects of Land Reform: Regression Results 5.1 Regression Specification Following the discussion of determinants of household division above, we use the following regression specification: DIVivt = βi +δt +β1 HSi,t−1 +β2 Li,t−1 +β3 OBv,t−k +β4 OBv,t−k ∗Li,t−1 +β4 Pv,t−k +β5 Pv,t−k ∗Li,t−1 +β6 Cit +ivt where DIVivt denotes a dummy for division, or lands lost owing to division, by household i located in village v in year t. The regressors include apart from household fixed effects and year effects, lagged household size HS, landownership L, measures of implementation of Operation Barga OB and the patta program P in the village lagged by a few (k) years,21 interactions of these with lagged land owned by the household, and a dummy C for whether the household owned land in excess of the legal ceiling. We run both logit and linear probability versions of this regression. Standard errors of residuals ivt are clustered at the village level. We also use a similar specification for dummies for whether a household bought or sold land. The theory indicated that the likelihood of buying or selling would be related to the ratio of (lagged) household size to land owned, so we include a specification where the log of this ratio is used as a single regressor apart from household and time effects. This corresponds to a different functional form, representing differences in endowment compositions that motivate land market transactions. 5.2 Household Division Table 8 presents a logit regression predicting the event that a household experienced a division in any given year. Columns 1 through 3 show the results for the full sample and 4 through 6 for the restricted sample. Columns 3 and 6 present the specification described above, with the other columns showing a more parsimonious specification which drop some of the land reform variables. Focusing on columns 3 and 6, we find growth in household size significantly raising the odds of household division, controlling for land owned. This confirms the notion that demographic growth was a key determinant of division. The effect of land owned itself was positive and insignificant, in contrast to the theoretical prediction (based only on purely economic reasons). The positive effect could be the result of anticipated land reform actions in the future which would be focused especially on large landowning households, thus motivating larger landowners to subdivide in order to evade such actions. This interpretation is confirmed by the significance of the above-ceiling dummy in columns 2 and 3 in the full 21 We take the average of these in three preceding years. sample, which causes the size of the land coefficient to halve in magnitude compared with column 1 and become insignificant. Columns 2,3,4,5 show that the effects of Operation Barga are as anticipated by the theory: division rates fell significantly f or s mall l andowning h ouseholds, a nd l ess s o for larger landowning households. The intercept and slope effects are significant at the 1% level in the full sample and either 5 or 1% in the restricted sample. For households owning more than four acres of land, the implied effect is positive. A separate regression interacting ‘bargaland’ with a dummy for households owning more than 4 acres of land confirmed the fact that Operation Barga induced a significant r ise i n d ivision r ates f or l arge landowning households. This suggests that the net effect on inequality would be negative. Columns 3 and 6 show effects of the patta program: a positive effect on division rates which is significant at 1%, and an interaction with land size which is negiligible and insignificant. Hence the land title program raised division rates uniformly, which is consistent with the theoretical predictions. The implied effect on land inequality is ambiguous (abstracting, of course, from the direct effect which would have reduced inequality owing to reduced landlessness). Table 10 shows a corresponding regression for the amount of land lost by a household as a consequence of division. While household size and land owned continue to be significant, the land reform variables are not significant. H ence t he m ain e ffects o f t he l and r eform on division was mainly through whether or not division occurred. A possible concern with the preceding regression is possible endogeneity of land reforms: their implementation rates may be correlated with time-varying village-level unobservables that affect division rates. To examine this issue, Table 9 presents instrumental variable as well as OLS estimates of the effects of Operation Barga, in a linear probability regression of division rates for small (owning less than 2.5 acres) and large (more than 2.5 acres) landowning households, controlling for the patta program, lagged household size and landownership besides household fixed e ffects a nd y ear d ummies. We u se t he s ame s et o f i nstruments for Barga implementation as in Bardhan and Mookherjee (2011), based on the hypothesis that implementation rates were affected in a nonlinear way by political competition between the Left Front and its key rival the Indian National Congress (INC) in local government elections. This hypothesis was proposed and tested empirically in Bardhan and Mookherjee (2010), which showed a number of important determinants of local political competition: the share of seats in the national Parliament held by the INC, average vote share differences between the two parties in preceding elections to the state legislature in the district in question, interacted with the share of seats held by the Left Front in the previous local government administration. Additional predictors include local infrastructure, rainfall and the price of rice. In order to predict cumulative Barga rates, these predictors are also cumulated over past years. The first s tage r egressions s how t hese i nstruments h ave moderate predictive power, with an F-ratio ranging between 8 and 9, but there is indication of underidentification as judged by the Kleibergen-Paap test. The Hansen test of overidentifying restrictions is comfortably passed, on the other hand. OLS results are provided in columns 1 and 3 for the two groups of landowners, with corresponding IV results in columns 2 and 4. Consistent with the results of the logit in the previous Table, Barga implementation rates significantly reduce division rates among small landowners and raise them among large landowners. The gap between the OLS and IV estimates is not large, with the test of endogeneity bias failing to reject the hypothesis of absence of endogeneity at any significance level below 0.2. This assures us that the OLS estimates of the effects of Operation Barga are reliable.22 Note, however, that the estimated effects of the patta program from the linear probability regressions in Table 9 differ considerably from the logit estimates in Table 8. This casts doubt on the reliability of the estimates of the effects of the patta program. The other point to note concerns the quantitative effect of the land reforms on the division rates implied by these estimates: they are not large, and small especially when compared to the effects of demographic growth. We estimate growth in population of native households would have been associated with an increase in household size by approximately 1.3 members in the absence of division.23 The logit regressions in Table 8 imply the estimated marginal effect of a rise in household size by one member on the odds of dividing is .046, while the marginal effects of Operation Barga and the patta program on division odds are -.023 and .089 respectively.24 This implies that adding 1.3 members to household size would imply a rise in division rates by approximately .059. In contrast, the cumulative effect of Operation Barga was to lower division rates by .0017, while the patta program raised them by .0017 with an aggregate effect of zero.25 The linear probability model IV estimates in Table 9 implies that the Barga and patta programs caused division rates for small landowners to fall by .005 and .028 respectively, while adding 1.3 members would cause them to rise by .134. 5.3 Land Market Transactions Table 11 presents a logit regression for the event that a household engaged in a land sale. There is no evidence of any significant effects of lagged household size, but lagged landownership clearly matters. Consistent with the theory, columns 1 and 4 show a negative effect of the (log of the) ratio of lagged household size to landownership. The corresponding re22 Unfortunately, we have not succeeded in finding a set of instruments that can help predict patta implementation, so we cannot test for endogeneity bias with regard to that program specifically. 23 We estimate native village population grew 22% between 1978 and 1998, using the number of village households from the indirect survey in Bardhan and Mookherjee (2006, 2010) and the average household sizes in these two years from the direct household survey. Applied to a mean household size of 5.7 in 1978, this implies household size would have expanded by approximately 1.3 members, in the absence of any division. 24 All marginal effects are calculated assuming zero values of fixed effects, and averaged across land sizes in the sample. 25 The observed cumulative value of bargaland in 1998 was .057 and pattaland was .067. gression using household size and land as separate regressors shows landownership rather than household size to be the important determinant. Columns 3 and 6 show that the only significant e ffect o f t he l and r eforms a ppeared w ith t he patta p rogram r aising l ikelihood of sales among large landowning households (consistent with the interpretation of increases in costs of hired labor and anticipated redistribution in the future). Operation Barga did not have a significant e ffect o n l ikelihood o f sales. Next, Table 12 presents the corresponding regression for likelihood of buying land. We see a converse negative effect of land owned. Operation Barga raises probability of purchase by small landowning households, and reduces it for large landowning households (e.g., those owning more than four acres). The patta program has significant effects only in the full sample, where it raises probability of purchase among those owning more than two acres. It is natural to expect some crowding out of purchases made by landless and marginal landowners owing to distribution of free pattas from the government. Most likely the observed effect purchases amongst the non-marginal owners arises from the sales induced for large landowners (as seen in the previous Table). However this result is not robust to the choice of sample. In summary, the evidence shows that the land reforms induced greater activity in the land market. Operation Barga seems to have induced transactions which would have caused inequality to fall. The evidence concerning the effects of the patta program is similar, but less clear. 5.4 Immigration Table 13 considers the determinants of village demographic share of post-1967 immigrants, in subsequent years. Columns 1 and 3 regresses the proportion of immigrants in the village sample, on contemporaneous average land per household in the village, and measures of land reform implemented so far, besides village and year dummies. Columns 2 and 4 present corresponding Arellano-Bond regressions with a lagged dependent variable. The first two columns use the proportion of land area covered by the two programs to measure program coverage, while the last two columns use proportion of the number of households residing in 1978 that were covered by the program after 1978. We use this latter measure because the demographic coverage of the programs may be more relevant to determining their impact on future immigration flows i nto t he v illage. E ven i f t he t otal a mount o f l and distributed to the poor may be small in relation to the amount of land in the village, a high probability of receiving a land title to a small plot of land may induce potential immigrants to settle in the village. Moreover, wage rates and employment opportunities for hired labor may be more sensitive to the demographic measure of program size, as land title recipients are likely to switch from supplying labor on the local market to working on their own plots. The demographic measure of program coverage normalizes using as base the number of households residing in 1978, rather than the current number of households, to avoid the possibility that the latter may be endogenously affected by the reforms through their effect on division rates. Hence the regressions reported in the third and fourth columns are run with post-1978 data. There is a significant n egative c oefficient o n l and a vailability, p ossibly r eflecting the greater tendency for immigrants to settle in areas of high population density (owing to their concern for finding e mployment a s w orkers). T he fi xed eff ect re gressions wi thout a lagged dependent variable in the first and third columns show contrasting effects of different measures of Barga program coverage. The area-based measures in column 1 show that Operation Barga implementation discouraged immigration (significant a t 1 0%), w hile the demographic coverage measure in column 3 encouraged immigration (significant a t 5%). Both measures of patta coverage, however, increased immigration (significant at 1 0%). These effects become statistically insignificant w hen a l agged d ependent v ariable i s i ncluded in the regression. Hence the results concerning effects on immigration are not robust to the specification. 5.5 Reduced Form Impact of Land Reforms on Land Inequality and Landlessness From the preceding results, we expect tenancy reforms to have had a negative effect on inequality via their impact on division and market transactions.26 The indirect effect of the patta program is less clear. Another notable difference is that unlike Operation Barga, the latter program will have a direct impact on lands owned by the poor, reducing landlessness and inequality. Table 14 presents a cross-village regression predicting 1998 inequality (measured either by the Gini coefficient or coefficient of variation) by the land reforms implemented since 1968, controlling for the level of inequality in 1968. Here the land title program (measured by the proportion of land area distributed) registers a significant negative coefficient, except in the Gini regression in the restricted sample. The predicted impact of the patta program on the coefficient of variation, for instance, ranges from a reduction of .019 to .053, which is approximately one-third to five-sixth o f t he o bserved i ncrease i n t his i nequality measure. However, the Barga implementation rate does not have a significant effect on the change in inequality. The significant i nequality r educing i mpact o f t he patta p rogram i s l ikely t o r esult from the direct effect of that program on landlessness. Approximately one in seven households in 1998, and one in five in 1978 had received a land title, so the direct impact on the proportion of landless population was sizeable. We have already seen from the decompositions that changes in landlessness was the principal factor driving rising inequality. To verify this interpretation, Table 15 runs a regression of the landlessness in 1998 on the proportion of households that were registered under Operation Barga and received land titles respectively, 26 Using the area-based measure of Barga implementation, the same is true for the impact of the program on immigration, which was negative (column 1 in Table 13), implying a negative impact on landlessness. controlling for landlessness in 1978. All variables are expressed as a proportion of the number of 1978 households, to avoid the problem arising from endogeneity of the the number of households with respect to the land reforms. This regression shows that the patta program had a significant n egative e ffect o n l andlessness, u nlike O peration B arga. H owever the regression coefficient is of the order of 0.1, indicating that titles distributed to 27% of the 1978 population resulted in a decline in landlessness by approximately 2.7% twenty years later. On the other hand, the coefficient on the 1978 landlessness ratio is 0.88, implying that the direct effect of the program (assuming it all happened in the year 1978) ought to have reduced landlessness in 1998 by 23.8%.27 Since the actual impact is so much lower, this indicates that the indirect effects of the patta program, if anything, served to increase landlessness. This could have happened owing to the effect of the program in raising immigration into the village, or raising division rates among small landowners (though the evidence here is less clear, owing to the conflicting r esults f rom d ifferent s pecifications of th e division regressions). In any case, we do not find e vidence t hat t he i ndirect e ffects o f e ither land reform program accounted for a significant r eduction i n l and i nequality. T he d irect effects of the patta program on reducing landlessness, on the other hand, helped reduce inequality. 6 Summary and Concluding Observations The main question addressed by this paper concerns the effectiveness of land reforms in reducing land inequality in the context of the West Bengal experience with these programs since the late 1960s. Measured in terms of proportion of land areas affected, the direct impacts of the land reforms in changing landownership patterns were negligible, compared with the effects of household division, market transactions and immigration. We subsequently estimated indirect effects of the reforms on the land distribution through their impact on divisions, market transactions and immigration. We found evidence that they did affect household division rates and market transactions in ways consistent with theoretical predictions. Yet these effects were quantitatively insignificant, r elative t o t he e ffects o f population growth arising from excess of births over deaths. The overall impact of Operation Barga had no significant i mpact o n o bserved c hanges i n l and i nequality o r l andlessness. The patta program did reduce landlessness and inequality somewhat, but by an extent that was substantially smaller than would have been expected from the direct impact of that program on landlessness. The indirect effects, if any, of the patta program caused landlessness to increase rather than decrease. The results of this paper indicate the need to focus on other determinants of the high rates of household division in future research. In particular, the causes of high rates of division 27 Of course the pattas were not all distributed in 1978 but later. Given the high degree of persistence in landlessness, the impact on the 1998 landlessness proportion would have been even higher. among small and marginal landowning households that induced rapid growth of landlessness need to be better understood. Did these arise from factors normally associated with later stages of demographic transition, wherein fertility rates drop among wealthier households but not to the same extent amongst poorer households, or mortality rates drop especially among poorer households? What was the role of intra-household heterogeneity in education and age, or other conflicts w ithin h ouseholds w ith r espect t o c ollective c onsumption goods (stressed by Foster and Rosenzweig (2002)), rising incomes and insurance opportunities and changing cultural norms which undermined the traditional joint family? Likewise, greater attention needs to be paid to determinants of land market transactions and immigration flows. We found evidence of an active land market, which served on the whole to equalize the land distribution. This suggests that fears regarding inequalityenhancing effects of deregulating land markets further may be unfounded. Immigration inflows seemed to respond positively to larger demographic coverage of land reforms, one of the reasons why landlessness may increase as an indirect outcome of these reforms. Besides, immigrants seem to be attracted into areas of high population density, suggesting that immigration tends to cause population concentrations in particular areas. A more thorough and careful analysis is needed to understand the determinants and implications of land markets and immigration flows to obtain a better understanding of land distribution changes as the West Bengal economy modernizes. Figure 1: Agricultural land per household, various measures (1967-2004) full .55 .5 .45 .4 .35 share a 4 3 2 1 0 agricultural 1970 1975 1980 1985 1990 1995 2000 year Average 50th 75th % restricted Note: gricultural of sample landless percentile of The landless land 25th sample inhouseholds acres percentile is not shown since it is equal to zero for the whole period analyzed. .55 .45 .5 share of landless households 3 1970 1975 1980 1985 1990 1995 2000 year .4 .35 1 0 agricultural land in acres 2 .4 .35 0 1 agricultural land in acres 2 .45 .5 share of landless households 3 .55 4 restricted sample 4 full sample 1970 1975 1980 1985 1990 1995 2000 year Average 50th percentile Average 50th percentile 75th percentile % of landless 75th percentile % of landless Note: The 25th percentile is not shown since it is equal to zero for the whole period analyzed. Figure 2: Household size, various measures (1967-2004) full h 8 7 6 5 4 3 household 1970 1975 1980 1985 1990 1995 2000 year Average 25th 50th 75th restricted ousehold sample percentile members sample 7 household members 5 6 4 3 3 4 household members 5 6 7 8 restricted sample 8 full sample 1970 1975 1980 1985 1990 year 1995 2000 1970 1975 1980 1985 1990 year 1995 2000 Average 25th percentile Average 25th percentile 50th percentile 75th percentile 50th percentile 75th percentile Figure 3: Agricultural land per capita, various measures (1967-2004) full .5 .45 .35 .3 .share 0 .6 .4 .2 agricultural 1970 1975 1980 1985 1990 1995 2000 year Average 50th 75th % restricted Note: 2 of sample landless percentile of The landless land 25th sample inindividuals acres percentile is not shown since it is equal to zero for the whole period analyzed. .5 .4 .45 share of landless individuals .3 .35 agricultural land in acres .2 .4 0 0 .3 .35 agricultural land in acres .2 .4 .4 .45 share of landless individuals .5 .6 restricted sample .6 full sample 1970 1975 1980 1985 1990 1995 2000 year 1970 1975 1980 1985 1990 1995 2000 year Average 50th percentile Average 50th percentile 75th percentile % of landless 75th percentile % of landless Note: The 25th percentile is not shown since it is equal to zero for the whole period analyzed. Figure 4: Immigration (1967-2004) .full 0 .4 .3 .2 .1 1970 1975 1980 1985 1990 1995 2000 year % restricted 1 of sample immigrants sample from Bangladesh .1 % of immigrants .2 .3 .4 restricted sample 0 0 .1 % of immigrants .2 .3 .4 full sample 1970 1975 1980 1985 1990 1995 2000 year 1970 1975 1980 1985 1990 1995 2000 year % of immigrants % of immigrants % of immigrants from Bangladesh % of immigrants from Bangladesh Figure 5: Agricultural land per capita, only natives (1967-2004) full .55 .5 .45 .35 .share 0 .6 .4 .2 agricultural 1970 1975 1980 1985 1990 1995 2000 year Average 50th 75th % restricted Note: 2 of sample landless percentile of The landless land 25th sample in acres percentile is not shown since it is equal to zero for the whole period analyzed. .5 .6 .35 .4 .45 share of landless agricultural land in acres .2 .4 0 0 .35 .4 .45 share of landless agricultural land in acres .2 .4 .5 .6 .55 restricted sample .55 full sample 1970 1975 1980 1985 1990 1995 2000 year 1970 1975 1980 1985 1990 1995 2000 year Average 50th percentile Average 50th percentile 75th percentile % of landless 75th percentile % of landless Note: The 25th percentile is not shown since it is equal to zero for the whole period analyzed. Figure 6: Agricultural land kernel densities, various years .full density .8 .6 .4 .2 0 1 2 a 3 agricultural 1970 1985 2000 restricted Notes: kernel 2gricultural sample function Landless land sample and households a bandwith areof excluded, 0.2. as well as households owning more than 3 acres of land. All graphs use the epanechnikov .6 density .4 .2 0 0 .2 density .4 .6 .8 restricted sample .8 full sample 0 1 1970 agricultural land 1985 2 3 2000 0 1 1970 agricultural land 1985 2 3 2000 Notes: Landless households are excluded, as well as households owning more than 3 acres of land. All graphs use the epanechnikov kernel function and a bandwith of 0.2. Figure 7: Average within-village land inequality, various measures (1967-2004) fullcoefficient 1.8 1.7 1.6 1.5 1.4 coefficient .66 .64 .62 .6 .58 .56 gini 1970 1975 1980 1985 1990 1995 2000 year Gini Coefficient restricted sample of of variation sample variation .66 1.6 coefficient of variation 1.7 .64 1970 1975 1980 1985 1990 year Gini coefficient 1995 .56 1.4 .58 1.5 1.7 1.6 coefficient of variation .56 1.4 .58 1.5 gini coefficient .6 .62 .66 .64 gini coefficient .6 .62 1.8 restricted sample 1.8 full sample 2000 1970 Coefficient of variation 1975 1980 1985 1990 1995 year Gini coefficient 2000 Coefficient of variation Figure 8: Proportion of landless households (1967-2004) .45 % of landless households .35 .4 .3 .25 .25 .3 % of landless households .35 .4 .45 .5 restricted sample .5 full sample 1970 1975 1980 1985 year 1990 1995 2000 1970 1975 1980 1985 year 1990 1995 2000 Figure 9: Average within-village land inequality, contribution to the Gini coefficient by channel (1967-2004) .full 0 -.02 .06 .04 .02 contribution 1970 1980 1990 2000 year Division/exit Market Reform restricted Note: The (Includes 02 sample restricted contribution contribution Eachof contribution sample line each landless sample represents channel corresponds and the immigrants) contribution to Sample of (2) each in channel Table 3,toi.e. the only change land history in the gini correct. coefficient. (Includes landless and immigrants) contribution of each channel 0 .02 .04 -.02 -.02 contribution of each channel 0 .02 .04 .06 restricted sample .06 full sample 1970 1980 year 1990 2000 1970 1980 year 1990 Division/exit contribution Division/exit contribution Market contribution Market contribution Reform contribution Reform contribution 2000 Note: Each line represents the contribution of each channel to the change in the gini coefficient. The restricted sample corresponds to Sample (2) in Table 3, i.e. only land history correct. Figure 10: Average within-village land inequality, contribution to coefficient of variation by channel (1967-2004) .full 0 -.1 .2 .1 contribution 1970 1980 1990 2000 year Division/exit Market Reform restricted Note: The (Includes 1 sample restricted contribution contribution Eachof contribution sample line each landless sample represents channel corresponds and the immigrants) contribution to Sample of (2) each in channel Table 3,toi.e. the only change land history in the coefficient correct. of variation. (Includes landless and immigrants) contribution of each channel 0 .1 -.1 -.1 contribution of each channel 0 .1 .2 restricted sample .2 full sample 1970 1980 year 1990 2000 1970 1980 year 1990 Division/exit contribution Division/exit contribution Market contribution Market contribution Reform contribution Reform contribution 2000 Note: Each line represents the contribution of each channel to the change in the coefficient of variation. The restricted sample corresponds to Sample (2) in Table 3, i.e. only land history correct. Figure 11: Land market: Sales and purchases per household (1967-2004) .02 land in acres 0 -.02 -.04 -.04 -.02 land in acres 0 .02 .04 restricted sample .04 full sample 1970 1975 1980 1985 year Sales 1990 1995 Purchases 2000 1970 Net sales 1975 1980 1985 year Sales 1990 1995 Purchases 2000 Net sales Figure 12: Average land reform implemented, official land records (1968-1998) As % of total households -.02 -.03 -.02 -.01 % of cultivable land 0 % of households -.01 0 .01 .01 .02 .02 As % of total cultivable land 1970 1975 1980 Barga 1985 year 1990 1995 1970 1975 Patta Source: West Bengal Block Land Reform Office (BLRO) for relevant villages. 1980 Barga 1985 year 1990 Patta 1995 Figure 13: Characterization of Stable Households without Land Market Transactions !" !" $%" $%" $&" !"#$%&' !"#$%&' $&" $%'" (""""""")""""""*" #" #" %+,-"3/"4567"2" %+,-"./"!01"2" Figure 14: The Effect of Incorporating Land Market Transactions !" (*' $%" !"#$%&' !" ()' (*' $%" +* ' $&" +-' +,' +' +)' ()' $&" #" #" '()" '*)" Notes: (a) Characterization of Stable Households with Land Market Transactions. (b) Land Transactions Induced by Expansion of Household Size. 38 Table 1: Comparing samples Full Sample (1) 5.159 (2.496) Land history Correct (2) 5.083 (2.380) Household size history correct (3) 5.228 (2.600) Both land and household histories correct (4) 5.098 (2.389) Diff. between columns (1) and (2) (5) 0.115 (0.071) Diff. between columns (1) and (3) (6) -0.051 (0.076) Diff. between columns (1) and (4) (7) 0.065 (0.074) Fraction of immigrant households 0.303 (0.460) 0.296 (0.457) 0.314 (0.464) 0.332 (0.471) -0.012 (0.012) -0.035 (0.013)** -0.046 (0.013)*** Total agricultural land 1.100 (2.265) 1.006 (2.167) 1.181 (2.361) 0.950 (2.081) 0.175 (0.062)** 0.019 (0.066) 0.208 (0.063)*** Irrigated agricultural land 0.732 (1.785) 0.658 (1.685) 0.792 (1.875) 0.618 (1.582) 0.136 (0.050)** 0.019 (0.053) 0.161 (0.051)** Unirrigated agricultural land 0.368 (1.287) 0.346 (1.261) 0.388 (1.315) 0.333 (1.219) 0.039 (0.033) 0.001 (0.034) 0.047 (0.034) % Landless 53.42 55.06 52.33 56.60 % Marginal (between 0 and 1.25 acres) 25.11 25.23 24.85 24.46 % Small (between 1.25 and 2.5 acres) 5.13 4.82 5.32 4.90 Household size % Medium (between 2.5 and 5 acres) 9.51 8.94 9.74 8.29 % Large (between 5 and 10 acres) 5.58 4.82 6.34 4.74 % Big (more than 10 acres) 1.26 1.12 1.42 1.01 N 2402 2099 1911 1697 Notes: Columns (1)-(4) report means with standard errors in parentheses. Means are computed using only survey answers for the year 2004. Column (2) includes those households for which the constructed land holding matched the reported in 2004, up to a 0.2 acres margin of error. Column (3) includes those households for which the constructed family size matched the reported in 2004, up to a 1 member margin of error. Column (4) include those households for which both the constructed land holding and family size matched the reported in 2004. Columns (5), (6) and (7) report tests for differences of means across column (1) and columns (2), (3), and (4), respectively. Robust standard errors are in parentheses. Tests are based on regressions with village fixed effects. Table 2: Changes in cultivable land and number of households, indirect survey Obs. Cultivable land in initial year Cultivable land in 1998 Change in cultivable land No. households in initial year No. households in 1998 Change in No. households 63 63 63 63 63 63 Cultivable land in initial year Cultivable land in 1998 Change in cultivable land No. households in initial year No. households in 1998 Change in No. households 26 26 26 26 26 26 Mean Std. Dev. Min Initial report prior to 1980 358.5 303.6 18.0 360.2 283.3 26.2 1.7 148.2 -843.8 231.0 219.5 24.0 419.5 380.3 47.0 188.5 192.6 -6.0 Initial report after 1980 230.6 170.1 4.6 217.6 149.2 9.6 -13.0 66.5 -225.3 236.7 156.0 18.0 346.7 186.9 60.0 110.0 83.0 10.0 Max 1265.5 1304.0 244.4 1083.0 1692.0 841.0 642.7 495.3 109.9 759.0 770.0 332.0 Notes: Cultivable land is measured in acres. 46 villages report cultivable land in 1977, 14 in 1978, 3 in 1979, 1 in 1980, 1 in 1981, 23 in 1983, and 1 in 1984. Table 3: Determinants of decrease in land holdings: cumulative changes at the household level, only natives (1967-2004) Sample: Land in 1967 Land change Lost due to land division Lost through sales Gained through purchases Lost due to reform Gained due to reform Lost as a gift Gained as a gift Lost for other reasons Gained for other reasons full 2.862 -1.370 -1.108 -0.557 0.467 -0.097 0.034 -0.116 0.030 -0.060 0.011 restricted 2.143 -0.926 -0.785 -0.475 0.373 -0.018 0.028 -0.097 0.032 -0.024 0.012 Notes: All numbers indicate average acres gained or lost per household. The category Lost for other reasons includes forced transfer, mortgaged, and lost due to natural disasters. All data comes from the household survey. Table 4: Changes in land distribution (restricted sample) Land Class Landless Marginal Small Medium Large Big Initial (1967-1971) 38.47 22.36 11.76 13.56 9.91 3.94 Final (2002-2004) natives and immigrants only natives 56.23 46.05 24.45 29.34 4.96 6.16 8.38 10.63 4.96 6.55 1.02 1.27 Notes: Numbers indicate the pecentage of households in each category. Table 5: Household division: Summary statistics (1) (2) (3) (4) Death of member of the household Exit of spouse of head Out-marriage Division due to other reasons # 250 18 365 3,940 Total 106.71 15.88 203.96 3,648.24 Mean 0.43 0.88 0.56 0.67 50th p. 0.00 0.03 0.01 0.06 75th p. 0.14 2.00 0.66 0.70 95th p. 3.00 2.82 2.00 3.16 Notes: All figures are in acres, except for #, the total number of events. Reasons stated in (4) include change in household size, change in income/expenditure, disputes, registration of tenants and threat of land reforms. Table 6: Proportion of households experiencing transactions, land reform and divisions (1967-2004) Sample: Sales Purchases Lost due to reform Gained due to reform Exits and division full 0.257 0.229 0.007 0.036 0.685 restricted 0.238 0.211 0.004 0.036 0.638 Notes: All numbers indicate the proportion of households with at least one event (sale, purchase, etc) between 1967 and 2004. Table 7: Division rates and proportion of land lost, in different size classes 1967-2004 (re- stricted sample) Land Class Landless Marginal Small Medium Large Big % of households 4.46 4.43 4.83 4.53 4.19 5.03 % land lost 0.21 1.07 1.52 1.19 1.05 1.51 Notes: The first column shows the annual proportion of households that divided in a given period of time. The second column indicates the proportion of land that households lost due to division. Division means one or more members left the household. Numbers are percentages. Table 8: Determinants of household division, using past reform, average of last three years (logit) Dep. Variable: Sample: Lagged HH size Lagged land Probability of fragmentation full (1) 0.4169*** (0.0282) 0.1250*** (0.0431) (2) 0.4480*** (0.0320) 0.0672 (0.0465) -0.2675*** (0.0529) 0.0736*** (0.0269) 1.2192** (0.4970) 40,621 0.0868 36,870 0.0908 bargaland Lagged land*bargaland ceiling pattaland Lagged land*pattaland Observations Pseudo R2 restricted (3) 0.4479*** (0.0320) 0.0676 (0.0471) -0.3108*** (0.0534) 0.0838*** (0.0201) 1.2164** (0.4969) 0.8499*** (0.1156) -0.0001 (0.1003) 36,870 0.0911 (4) 0.4962*** (0.0373) 0.1166*** (0.0401) (5) 0.5207*** (0.0401) 0.0673 (0.0530) -1.0433** (0.4830) 0.2585** (0.1042) 0.5377 (0.6889) 28,011 0.0938 25,442 0.0973 (6) 0.5205*** (0.0401) 0.0674 (0.0529) -1.6901*** (0.6000) 0.3937*** (0.1196) 0.4892 (0.6662) 1.4246*** (0.3007) 0.1118 (0.2557) 25,442 0.0980 Notes: Logit coefficients reported with robust standard errors in parentheses, adjusted for clustering on villages. All regressions include year dummies and household fixed e ffects. T he variables p attaland and bargaland are computed as the sum over the previous three years of the share of land affected by each program over the total cultivable land in each village, using official land records. ( *** p<0.01, ** p<0.05, * p<0.1) Table 9: Determinants of household division, instrumenting for barga (linear probability model, full sample) Dep. Variable: Sample (landowners): Model: Barga (cumulative) Patta (cumulative) Lagged HH Size Lagged land Constant Observations Number of households R-squareda Kleibergen-Paap under-id test ( p-value) Hansen’s J over-id test ( p-value) Endogeneity test for Barga (p-value): First stage F-test First stage F-test (excl. inst.) Probability of division small large OLS (1) -0.1223*** (0.0123) -0.0652 (0.1808) 0.1030*** (0.0095) 0.0484 (0.0653) -0.4337*** (0.0424) 5,784 1,431 0.0917 IV (2) -0.0828*** (0.0313) -0.4246** (0.2138) 0.1027*** (0.0094) 0.0583 (0.0647) 5,677 1,324 0.0935 0.9959 0.6006 0.2002 9.03 8.17 OLS (3) 0.2184*** (0.0195) 0.2161 (0.2251) 0.0797*** (0.0107) 0.0045 (0.0034) -0.4938*** (0.0561) 1,590 429 0.0804 IV (4) 0.2547** (0.0990) 0.1604 (0.2270) 0.0752*** (0.0104) 0.0061 (0.0038) 1,547 386 0.0777 0.9968 0.6169 0.4120 9.61 9.54 Notes: Robust standard errors in parentheses, adjusted for clustering on villages. All regressions include year dummies and household fixed effects. The variables pattaland and bargaland are computed as the cumulative percentage of total land, using official land records. Small landowners are households with less than 2.5 acres of cultivable land. Barga is instrumented using interactions of the cumulative share of GP left share, cumulative % of INC seats in parliament, and the cumulative lagged average vote share difference in the district, as well as the number of households, rainfall, GP local irrigation and road expenditures, log price of rice, canals and roads in district, as described in Bardhan and Mookherjee (2011). a Adjusted R-squared in (1) and (3), and centered R-squared in (2) and (4) ( *** p<0.01, ** p<0.05, * p<0.1) Table 10: Determinants of land lost by households due to division Dep. Variable: Sample: Lagged HH size Lagged land full (1) 0.0159*** (0.0034) 0.0505*** (0.0095) Land lost due to division in acres restricted (2) 0.0187*** (0.0037) 0.0397** (0.0156) pattaland bargaland -0.0025 (0.0072) Lagged land*pattaland Lagged land*bargaland ceiling Constant Observations R-squared Number of households -0.1477*** (0.0316) 58,765 0.0305 2,304 -0.0015 (0.0060) 0.1755 (0.1216) -0.1901*** (0.0357) 54,175 0.0272 2,268 (3) 0.0185*** (0.0037) 0.0394** (0.0155) -0.1941 (0.1680) -0.0018 (0.0074) 0.0956 (0.0867) -0.0022 (0.0063) 0.1721 (0.1206) -0.1881*** (0.0357) 54,175 0.0283 2,268 (4) 0.0089** (0.0038) 0.0688*** (0.0190) (5) 0.0130*** (0.0035) 0.0378*** (0.0071) -0.0074 (0.0079) -0.1198*** (0.0266) 41,536 0.0348 1,681 0.0025 (0.0079) 0.0913 (0.1096) -0.1338*** (0.0217) 38,190 0.0187 1,649 (6) 0.0132*** (0.0035) 0.0366*** (0.0073) -0.1658 (0.2292) -0.0076 (0.0079) 0.2146 (0.2594) 0.0026 (0.0079) 0.0749 (0.1096) -0.1321*** (0.0220) 38,190 0.0217 1,649 Notes: OLS coefficients reported with robust standard errors in parentheses, adjusted for clustering on villages. All regressions include year dummies and household fixed effects. The variabes pattaland and bargaland are computed as the sum over the previous three years of the share of land affected by each program over the total cultivable land in each village, using official records. ( *** p<0.01, ** p<0.05, * p<0.1) Table 11: Determinants of land sales, using lagged land reform, average for past 3 years (logit) Dep. Variable: Sample: Pr(land sale) full (1)  log Lagged HH Size Lagged land (2) restricted (3) -0.9077*** (5) (6) -0.0199 (0.0467) 0.4801*** (0.0665) -0.0199 (0.0503) 0.4821*** (0.0771) -2.4925 (1.7640) -0.0077 (0.8096) 0.3344* (0.1807) -0.0849 (0.3718) 0.3708 (0.4442) 7,837 0.0823 -1.0861*** (0.1843) Lagged HH Size (0.1606) 0.0380 (0.0387) 0.1303** (0.0527) Lagged land pattaland bargaland Lagged land*pattaland Lagged land*bargaland ceiling Observations Pseudo R2 (4)  12,612 0.0639 12,612 0.0517 0.0377 (0.0429) 0.1271** (0.0503) -0.8769 (1.0416) 0.0281 (0.5598) 0.4335** (0.1854) -0.1083 (0.2508) 0.3170 (0.3398) 11,483 0.0504 8,623 0.0805 8,623 0.0829 Notes: Robust standard errors in parentheses, adjusted for clustering on villages. Regressions include household fixed effects and year dummies. pattaland and bargaland are computed as the sum over the previous three years of the share of land affected by each program over the total cultivable land in each village. ( *** p<0.01, ** p<0.05, * p<0.1) Table 12: Determinants of land purchases, using lagged land reform, average for past 3 years (logit) Dep. Variable: Sample: Pr(land purchase) full (1)  log Lagged HH Size Lagged land (2) restricted (3) 0.0626** (5) (6) 0.0027 (0.0482) -0.2485*** (0.0553) -0.0103 (0.0477) -0.2274*** (0.0564) 0.4786 (3.0567) 0.4823*** (0.0781) -0.0535 (0.8874) -0.1435*** (0.0446) -0.0039 (0.5406) 6,339 0.0572 0.0542 (0.0277) Lagged HH Size (0.0356) 0.0161 (0.0383) -0.2137*** (0.0522) Lagged land pattaland bargaland Lagged land*pattaland Lagged land*bargaland ceiling Observations Pseudo R2 (4)  10,438 0.0394 10,438 0.0532 0.0123 (0.0384) -0.2523*** (0.0517) -1.4950 (1.3523) 0.4894*** (0.0808) 0.7145*** (0.2504) -0.1207** (0.0562) 0.9256** (0.4715) 9,546 0.0519 6,952 0.0459 6,952 0.0600 Notes: Robust standard errors in parentheses, adjusted for clustering on villages. Regressions include household fixed effects and year dummies. pattaland and bargaland are computed as the sum over the previous three years of the share of land affected by each program over the total cultivable land in each village. ( *** p<0.01, ** p<0.05, * p<0.1) Table 13: Determinants of immigration, village panel (restricted sample) Dependent variable: Proportion of post-1967 immigrant households in village (1) (2) (3) (4) Fixed Arellano Fixed Arellano Effects Bond Effects Bond Lagged share of immigrants 0.4277*** 0.6266*** (0.1196) (0.0897) Lagged land per capita -0.0980* 0.0162 -0.1423** -0.0688** (0.0528) (0.0269) (0.0686) (0.0272) Lagged bargaland -0.0024* -0.0001 (0.0013) (0.0006) Lagged pattaland 0.0297* 0.0063 (0.0152) (0.0103) Lagged bargahh 0.0136** -0.0010 (0.0068) (0.0024) Lagged pattahh 0.0190* -0.0039 (0.0099) (0.0096) Constant 0.0822*** 0.1358*** 0.2246*** 0.0927*** (0.0273) (0.0325) (0.0266) (0.0233) 0.4499 0.3928 Adjusted R-squared p-value from test for f.o. autocorr. 0.0000 0.0000 p-value from test for s.o. autocorr. 0.5990 0.1922 Observations 2,666 2,577 1,869 1,869 No. villages 89 89 89 89 Notes: Robust standard errors in parentheses, adjusted for clustering on villages. Regressions include village and year fixed effects. In (1) and (2), pattaland and bargaland are computed as the proportion of land affected by each program over the total cultivable land in each village. In (3) and (4), pattahh and bargahh are computed as the proportion of households affected by each program over the total number of 1978 households per village. For these columns only data post-1978 is used. All measures of patta and barga are computed on the basis of official land records. ( *** p<0.01, ** p<0.05, * p<0.1) Table 14: The effect of land reform on inequality: cross section village regressions full sample Dependent variable: gini coefficient in 1968 Gini in 1998 (1) 0.6520*** (0.1259) coefficient of variation in 1968 cumulative bargaland cumulative pattaland Constant Adjusted R2 Observations 0.0024 (0.0042) -0.0630** (0.0302) 0.2660*** (0.0708) 0.3438 89 restricted CV in 1998 (2) 0.7559*** (0.1352) 0.0283 (0.0187) -0.3835** (0.1468) 0.5590*** (0.1629) 0.4176 88 Gini in 1998 (3) 0.6577*** (0.1247) -0.0176 (0.0149) -0.1025 (0.1231) 0.2965*** (0.0797) 0.5633 88 CV in 1998 (4) 0.8974*** (0.1367) 0.0388 (0.0521) -1.0571** (0.4879) 0.5388*** (0.1655) 0.4417 83 Notes: Robust standard errors in parentheses. Cumulative pattaland and bargaland are computed as the share of total land affected by each program over the total cultivable land in each village. ( *** p<0.01, ** p<0.05, * p<0.1) Table 15: The effect of land reform on the proportion of landless households: cross section village regressions (full sample) Dependent variable: 1998 landless households as a proportion of 1978 households proportion landless in 1978 0.894*** (0.071) barga registration as proportion of 1978 hhs 0.065 (0.040) pattas distributed as proportion of 1978 hhs -0.104** (0.035) Constant 0.108*** (0.029) Adjusted R2 .71 Observations 89 Notes: Robust standard errors in parentheses. Dependent variable is number of 1998 landless households divided by number of 1978 households in each village. Regressors are corresponding number of housholds landless in 1978, numbers registered under Operation Barga and given pattas between 1978 and 1998 respectively, divided by number of 1978 households. ( *** p<0.01, ** p<0.05, * p<0.1)