TY - JOUR
AU - Beg, Sabrin
AB - Abstract I test the land and labor market effects of a property rights reform that computerized rural land records in Pakistan, making digitized records and automated transactions accessible to agricultural landowners and cultivators. Using the staggered roll-out of the program, I find that while the reform does not shift land ownership, landowning households are more likely to rent out land and shift into non-agricultural occupations. At the same time, cultivating households have access to more land, as rented in land and overall farm size increase. I construct measures of farmer-level total factor productivity (TFP) and marginal product of land, and demonstrate evidence of improved allocative efficiency as land is redistributed toward more productive farmers. Aggregate district-level production data suggest a reduction in the dispersion of marginal products of land and an improvement in productivity. The results have implications for both the allocation of land across farmers and the selection of labor into farming, demonstrating that agricultural land market frictions present a constraint to scale farming and structural change in developing countries. Teaching Slides A set of Teaching Slides to accompany this article are available online as Supplementary Data. 1. Introduction Agricultural productivity growth is imperative for development and structural change. Developing countries, however, lag severely in aggregate agricultural productivity, despite the availability of modern and mechanized inputs.1 Recent literature argues that misallocation of factors of production contributes to productivity differences across countries. An emerging agenda for development economists is thus to examine the causes of misallocation for unpacking agricultural productivity lags in the developing world. Weak property rights and tenure insecurity lead to high transaction costs and market constraints that hinder the optimal allocation of productive inputs. Land market frictions thus impede the efficient trading of land and the occupational choices of individuals (de Janvry, Fafchamps, and Sadoulet 1991; Adamopoulos et al. 2017; Chen 2017). Agricultural landowners facing restrictions in renting out or selling their land choose to farm when it might be optimal to practice a non-agricultural activity. Relatedly, barriers to purchasing or tenancy prevent productive farmers from expanding the scale of operation and realizing returns to scale and mechanization. As a result, farms in lower middle income countries, including Pakistan, are small, unmechanized, and lag in productivity (Foster and Rosenzweig 2017). Moreover, about a third to half of Pakistan’s labor force works in agriculture, which constitutes only 15% of the total GDP (World Bank 2013). This paper establishes a causal link between tenure security and market activity, and the consequent implication for allocative efficiency, farm scale, and productivity. Property rights and land market transactions are non-existent or excessively informal in the vast majority of developing countries. In the context of this study (Punjab, Pakistan), land records have been maintained under the same structure since the colonial period—paper-based records of millions of landowners were held by 8,000 local officers or patwaris, who manually updated and managed these records.2 The inefficient and dispersed land records system has led to tenure insecurity, with owners relying on the discretion of the patwaris for any transaction or proof of ownership and tenancy rights. These barriers to land transactions and security of property result in low mobility of land, affecting land use and labor market choices of rural landowners. In 2009, the Punjab government launched the Land Records Management Information System to formalize and centralize land records in the province. Through this program, which was phased out in stages across all districts of the province, land records were obtained from the patwaris, computerized, and made available to the public at a service center in each subdistrict. While no titles were given out as part of the program, an owner or tenant can go to the designated center and obtain a government-attested copy of his ownership or tenancy status, implying improved access to land records and security of rights due to the program. All land transactions and changes to ownership or tenancy are conducted digitally at this designated center. The program thus represents an overhaul of an informal system that is replaced with a more centralized and computerized system. I use the staggered rollout of the program between 2011 and 2015 to document effects of the program. Specifically, I exploit variation in the timing of program start in any district and the share of program subdistricts within a district to identify causal effects. I test the program’s effect on rental market participation and labor choices of landowning households, on allocation of land across farmers, and on farm operation, particularly, farm scale, input usage, and productivity. To validate the identification strategy, I conduct tests to ensure early and late program districts do not demonstrate differential prior levels or trends in the main outcome variables, underlying soil quality and productivity, or macro-economic indicators. These tests confirm that program timing or intensity is unlikely to be driven by pre-existing differences across districts. I find that the program increased rental market transactions, as landowners are more likely to rent out land. Consistent with higher rental activity, the rate of agricultural participation by landowners declines, supporting the significance of market frictions in affecting the selection of workers across sectors. Landowning households shift into non-agricultural occupations, particularly business ownership. This increase in renting out is driven by lower income households, who are more likely to face tenure insecurity and market constraints. I do not find any significant effects on land ownership or land sales and purchases, suggesting the market frictions for land sales are higher, or that renting and selling are possible substitutes. While some landowners rent out land and exit agriculture, households that continue to cultivate increase the scale of farming as shown by more rented in land and higher average farm size in program districts. I rank farmers by productivity based on farm-level TFP calculated using detailed information on farm output and inputs. I find that higher TFP farmers in a district have greater farm land (and lower marginal product of land) after the program relative to low TFP farmers. Additionally, the dispersion in marginal products of land within a district is lower after the program. These findings support the hypothesis that market activity due to the program results in a more efficient allocation of land. Suggestive evidence of improved input usage and investment supports the scale and allocation effects of the program. Cultivating households are marginally more likely to switch crop choice and use pesticides. The program has no effect on average farm-level yield, but a positive effect on two different measures of aggregate productivity. Remote sensing data on vegetation across subdistricts are used as a proxy for crop production; I find a significant increase in the vegetation index as the program is rolled out. Additionally, district-level data on aggregate output by crop show greater improvements in cereal yield due to the program (significant at the 10% level). Taken together, the changes to land allocation, farming scale and inputs, and aggregate and remote sensing measures of agricultural output all suggest allocative efficiency and productivity improvement due to the program. I test the robustness of the main findings in a number of ways. In additional specifications, I adjust the control variables and sample years, control for simultaneous macro-economic trends, and drop the early program districts that may be subject to selection bias. I complement the main findings using alternate identification strategies, including a “stacked” difference-in-difference (DD), a standard timing DD using just timing variation, and an event study analysis. The results are highly stable across the various robustness specifications and strategies. Misallocation in the industrial sector is documented by Hsieh and Klenow (2009) and in agriculture by Restuccia and Santaeulalia-Llopis (2017) and de Janvry, Fafchamps, and Sadoulet (1991). The documentation of misallocation of land and capital across farming entities is supported by parallel research noting the dramatic differences in the scale of farming across countries (Adamopoulos and Restuccia 2014).3 Adamopoulos et al. (2017) demonstrate that aggregate agricultural productivity depends not just on the allocation of land and capital across farmers, but also on allocation of workers across sectors—in particular, the type of farmers who operate in agriculture (selection). Much of the literature is focused on the extent and consequences of misallocation, and less on the sources. Existing papers highlight the role of markets in allocative efficiency by using theoretical arguments or by demonstrating a correlation between market activity and misallocation.4 As market activity is endogenous, it is challenging to identify its role in factor allocation. Chen, Restuccia, and Santaeulàlia-Llopis (2021) use variation in the degree of land certification in Ethiopia to show that land rentals are associated with lower misallocation and higher agricultural productivity. Chari et al. (2020) demonstrate that legalizing land rentals in China improves the allocation of land across farmers and boosts aggregate productivity. In both these contexts, land is communally or state owned, and therefore the status quo is characterized by a lack of any land market. By demonstrating improved allocative efficiency and productivity as a result of legalizing land rentals, these papers provide a justification for private property rights. However, even with private property rights, tenure insecurity can be high under informal or partially enforceable rights and contracts causing significant market frictions. This is apparent in the context of Punjab, where only 21% of landowners lease their land and over 80% of farms are under 10 acres. Theoretical work on property rights testifies to the role of tenure security on resource allocation. Besley and Ghatak (2010) identify two broad channels through which property rights affect allocation: (1) limiting expropriation; and (2) facilitating market transactions.5 Empirically, the positive effects of land titling and certification programs on “limiting expropriation” and incentivizing investment are well documented.6 Less consistent evidence has been documented for the theoretical argument that tenure security facilitates market activity. Field and Torero (2006), Do and Iyer (2008), and Galiani and Schargrodsky (2010) do not find that titling significantly improves credit access, while Wang (2012), Carter and Olinto (1996), and López and Romano (2000) argue that that they do. Deininger and Goyal (2012) find that land registry computerization in India increases credit access, though the effects are modest and only in urban areas. The existing literature lacks comprehensive evidence of how tenure insecurity affects land rental and sales in particular.7 Even fewer papers systematically identify the effect of property rights and security on labor choices, particularly in rural areas.8 This paper fills this gap in the property rights literature by documenting the benefits of a land rights computerization program in progressing tenure security, and facilitating land rental and labor market allocation. Specifically, I make two major contributions to the extensive body of empirical literature on property rights and misallocation. First, I provide direct evidence of the role of property rights insecurity in hindering agricultural land rental. I build on former work by demonstrating frictions in land rental activity even with privately owned property. The second contribution of my paper is the additional effects that I document on labor allocation of landowning households as rental transaction costs go down. These contributions depart from the focus of the existing property rights literature on their effect on investment incentives, and are complementary to the broader literature on property rights institutions and agricultural productivity (Bellemare 2013; Newman, Tarp, and Van Den Broeck 2015; Gottlieb and Grobovšek 2019). These findings also contribute to understanding the process of structural change and urbanization in the context of South Asia (Binswanger-Mkhize 2013). Agricultural participation is still considerably high in South Asian countries—approximately 50% of the total labor force in India, Pakistan, and Bangladesh (compared to 24% for middle income countries) (World Bank 2013). On the other hand, agriculture accounts for just 18% of the GDP on average for South Asia.9 Improving tenancy security and rights of land use can stimulate labor market allocation and structural transformation. The paper further highlights that Information and Communication Technology (ICT) in governance and public service delivery holds substantial promise for lower income nations with limited state capacity (Banerjee and Jain 2003; Ghosh and Banerjee 2006), and contributes to the literature on the positive impacts of digitization broadly on productivity and development (Bresnahan, Brynjolfsson, and Hitt 2002; Bloom et al. 2014). The next section describes the background of land records in Punjab and the Land Record Management and Information Systems program. Section 3 describes the data and empirical specification for the main results, and Section 4 describes the results and mechanisms. Section 5 discusses the validity of findings and offers additional robustness checks. Finally, Section 6 concludes. 2. Background10 2.1. Agriculture and Land Records in Punjab Punjab, the context of the study, is the most populated province of Pakistan with 80.5 million inhabitants (55.6% of the country’s population), 70% of whom live in rural areas. The Board of Revenue bears responsibility for the administration of agricultural land, which is mostly privately owned. The history of the land revenue system in Pakistan dates to pre-colonial rulers who introduced a system of land administration, which was improved and formalized by the British colonial government and then underwent minimal changes over a 60-year period after Pakistan’s independence. Several levels of administration are involved in land record maintenance: District, Subdistrict, Kanungo Circle, and Patwar Circle. Patwaris, or the local officers at the Patwar Circle level, are the custodians of land rights records—in Punjab, about 8,000 patwaris maintain paper-based land records pertaining to 20 million land owners, at times holding them in cloth bags. Among various land record statements described in further detail in the Online Appendix, the most relevant is the “Land Right Holders Register”, which lists the owners of each land parcel demarcated in a corresponding cadastral map of each village. Any changes to land rights are recorded in a separate register of mutations, which is used to update the register of right holders every four years. Tenants’ and landowners’ rights, as well as updates that arise due to rental or sale, are thus recorded at the discretion of the patwari and revenue officers above him in the bureaucratic chain. The manual and decentralized system is potentially prone to corruption and mistakes, lowering tenure security for owners and cultivators. A survey conducted by Gallup Pakistan for the Board of Revenue found that 42% of a sample of land owners and cultivators from Punjab report higher dissatisfaction with the system of land records than with other government departments. A total of 64% of farm households describe the system as lacking transparency, and 82% report ever having to pay a bribe to obtain land record services. A total of 76% of respondents in the poll reported illegal occupation of land as the main form of land dispute, and 56% identified that the major source of all land disputes was incorrect land records. Land transactions are uncommon. In the Pakistan Rural Household Survey (2001), 87% of landowners either inherited their land or obtained it from the government. Universal certification or titling is not prevalent, though official documents can be obtained based on verification by land revenue officials, albeit through a lengthy process. A request for obtaining a title begins at the patwari level and goes up the bureaucratic chain to the revenue office. Land ownership is verified by the revenue office through correspondence with the patwari who locates and confirms the rights of the landholder in his manual records. After verification, senior revenue officials issue a title to the landowner. Among the rural landowning household sample from 2001, only 45% have a “fard” (title) or an ownership document on a registered stamp paper for their property. Of the owners with title documents, only 25% report not having to submit payment to a revenue official beyond the legal title registration fee.11 Even for those with titles, tenure security may be low as land records are dispersed and not easily accessible or verifiable. Of the households, 11% report they cannot sell their land if they wanted to. In 2010, with a similar land administration system, the Government of Punjab in India made attempts to abolish colonial posts like patwaris who were often accused of corruption and making “fraudulent changes” in revenue records under their jurisdiction (Sural 2013). 2.2. Land Record Management and Information Systems Program Beginning in the years 2005–2009, the Government of Punjab received financial support from the World Bank to begin the computerization of land records to improve service delivery and enhance the perceived level of tenure security. The main objective of this endeavor was to facilitate increased access to land records at low costs, specifically for the poorest and least-connected households. The provincial government department noted that: Inequalities of land distribution, tenure insecurity and difficulties associated with the land administration and registration system are closely interrelated and continue to impose significant constraints on both rural and urban populations, particularly the poor. Land transactions are relatively high cost, and disputes about accuracy of land rights are caused, among others, by the inefficient and dispersed land records system. As a result land markets are thin and land prices are in excess of the discounted value of potential agricultural earnings from land. (World Bank–Project Information Document 2005) The first objective of the program was to computerize all rural land records, including the list of land right holders (owners and cultivators), as maintained by the patwaris. The second objective was to establish a service center in each subdistrict of the province to host these records and to replace the lower-level land record officers for maintaining and updating these records, and providing citizens with land mutation, title issuance, and other land record related services directly. The computerized records establish both the identity of the owner and tenant, and can be located on the internet or obtained from the designated service centers. The right holders (owners or tenants) can visit a service center where the staff can use their national ID number to search and verify their record, providing the client with a government-attested copy within minutes. Any mutation, due to to sale, transfer, or inheritance, is to be registered at the same service center. Approximately 150 centers across the province now provide automated land records services, reducing the average time required to complete transactions from 2 months to 45 minutes (Gonzalez 2016). While the service centers increased access to digitized records of ownership and cultivation rights, they may have increased distance to land records. Initially, a patwari was available for each patwar circle, which comprises a few proximate villages, and was well-known to all village members in his jurisdiction—once all service centers are fully operational, only one center is available per subdistrict.12 Though it may seem individuals face higher travel costs to access the centralized records, transacting parties were required to visit district revenue offices at several stages of any transaction and thus incurred high travel costs even prior to the reform. As the service centers provided all land transaction services at one location, the time and distance costs could effectively be lower after the program even with fewer service centers than patwaris. Changes to cultivation, for instance in the case of land rental, are still initially reported to the patwari, who then sends updated records to service centers at the beginning of each agricultural season. Rental transactions thus do not entail the travel costs, but the records and rights are transitioned from being manual and disaggregated to digital, central, and verifiable. The program thus resulted in two main changes to the pre-existing system: (1) centralized record keeping for ownership and tenancy rights; and (2) low cost and centralized land transactions including access to title documents. By making the computerized land records centrally available at a subdistrict level, the new system decreased the influence of the local officers and patwaris and can have potential effects on tenure security of owners and tenants, and consequently on the land market. 3. Empirical Strategy and Data 3.1. Data Program Rollout The program data are obtained from the Board of Revenue of Punjab, outlining the operational date for each subdistrict level service center in the province. Using the district boundaries from the pre-program period, there are 34 districts and 143 subdistricts in total.13 All 143 land records service centers opened between 2011 and 2015. Figure 1 shows the rollout of the service center openings and Figure 2 shows the number of subdistrict centers opening in each year. I construct ProgramIntensity|$_{dt}$| in any year t and district d as the share of subdistricts in d that have received the program by year t. Online Appendix Table A.1 shows the average values of ProgramIntensity|$_{dt}$| over the sample period. The government sought to roll out the service centers in no specific order though not strictly randomly, and the proposed identification strategy will alleviate any selection bias. Figure 1. Open in new tabDownload slide Program rollout. Figure 2. Open in new tabDownload slide Program openings by year. Household Outcomes The household data are obtained from Household Income and Expenditure surveys (HIES), conducted bi-yearly across the country—I use five HIES survey rounds from 2005 to 2015. These surveys, conducted in 2005–2006, 2007–2008, 2011–2012, 2013–2014, and 2015–2016, collect demographics, employment, expenditure, and saving information from a repeated cross-section of approximately 6,600 (3,800 rural) households from Punjab in each round.14 Thus, the data set has 19,067 rural households across 5 data rounds, and I focus on agricultural households. Specifically, I report outcomes for landowners (households that own agricultural land) and cultivators (households that operate a farm). There are 7,597 landowning households and 7,256 cultivating households.15 Summary statistics from the household data used for the analyses are shown in Table 1. Table 1. Summary statistics for households. Variables . (1) . Panel A: landowning households (N = 7,597)  Household rents out agricultural land 0.21 (0.405)  Household cultivates a farm 0.81 (0.395)  Household cultivates own farm 0.62 (0.484)  Household member works in agricultural land 0.81 (0.392)  Household share agricultural income 0.65 (0.391) Panel B: cultivating households (N = 7,256)  Total farm area cultivated (acres) 6.17 (8.070)  Land rented in on cash rent (Y/N) 0.25 (0.433)  Area rented in on cash rent (acres) 6.33 (10.27)  Land rented in on sharecropping (Y/N) 0.07 (0.253)  Area rented in on sharecropping (acres) 7.56 (14.49)  Output (value) per acre 55.63 (37.99)  Profit (output value − expenses) per acre 32.07 (24.38)  Grows wheat (Y/N) 0.91 (0.281)  Grows rice (Y/N) 0.31 (0.461)  Grows maize (Y/N) 0.07 (0.257)  Grows cotton (Y/N) 0.34 (0.472)  Grows sugarcane (Y/N) 0.17 (0.373) Variables . (1) . Panel A: landowning households (N = 7,597)  Household rents out agricultural land 0.21 (0.405)  Household cultivates a farm 0.81 (0.395)  Household cultivates own farm 0.62 (0.484)  Household member works in agricultural land 0.81 (0.392)  Household share agricultural income 0.65 (0.391) Panel B: cultivating households (N = 7,256)  Total farm area cultivated (acres) 6.17 (8.070)  Land rented in on cash rent (Y/N) 0.25 (0.433)  Area rented in on cash rent (acres) 6.33 (10.27)  Land rented in on sharecropping (Y/N) 0.07 (0.253)  Area rented in on sharecropping (acres) 7.56 (14.49)  Output (value) per acre 55.63 (37.99)  Profit (output value − expenses) per acre 32.07 (24.38)  Grows wheat (Y/N) 0.91 (0.281)  Grows rice (Y/N) 0.31 (0.461)  Grows maize (Y/N) 0.07 (0.257)  Grows cotton (Y/N) 0.34 (0.472)  Grows sugarcane (Y/N) 0.17 (0.373) Notes: Data are from the HIES surveys. Landowning households report agricultural land ownership; cultivating households report farming agricultural land. Open in new tab Table 1. Summary statistics for households. Variables . (1) . Panel A: landowning households (N = 7,597)  Household rents out agricultural land 0.21 (0.405)  Household cultivates a farm 0.81 (0.395)  Household cultivates own farm 0.62 (0.484)  Household member works in agricultural land 0.81 (0.392)  Household share agricultural income 0.65 (0.391) Panel B: cultivating households (N = 7,256)  Total farm area cultivated (acres) 6.17 (8.070)  Land rented in on cash rent (Y/N) 0.25 (0.433)  Area rented in on cash rent (acres) 6.33 (10.27)  Land rented in on sharecropping (Y/N) 0.07 (0.253)  Area rented in on sharecropping (acres) 7.56 (14.49)  Output (value) per acre 55.63 (37.99)  Profit (output value − expenses) per acre 32.07 (24.38)  Grows wheat (Y/N) 0.91 (0.281)  Grows rice (Y/N) 0.31 (0.461)  Grows maize (Y/N) 0.07 (0.257)  Grows cotton (Y/N) 0.34 (0.472)  Grows sugarcane (Y/N) 0.17 (0.373) Variables . (1) . Panel A: landowning households (N = 7,597)  Household rents out agricultural land 0.21 (0.405)  Household cultivates a farm 0.81 (0.395)  Household cultivates own farm 0.62 (0.484)  Household member works in agricultural land 0.81 (0.392)  Household share agricultural income 0.65 (0.391) Panel B: cultivating households (N = 7,256)  Total farm area cultivated (acres) 6.17 (8.070)  Land rented in on cash rent (Y/N) 0.25 (0.433)  Area rented in on cash rent (acres) 6.33 (10.27)  Land rented in on sharecropping (Y/N) 0.07 (0.253)  Area rented in on sharecropping (acres) 7.56 (14.49)  Output (value) per acre 55.63 (37.99)  Profit (output value − expenses) per acre 32.07 (24.38)  Grows wheat (Y/N) 0.91 (0.281)  Grows rice (Y/N) 0.31 (0.461)  Grows maize (Y/N) 0.07 (0.257)  Grows cotton (Y/N) 0.34 (0.472)  Grows sugarcane (Y/N) 0.17 (0.373) Notes: Data are from the HIES surveys. Landowning households report agricultural land ownership; cultivating households report farming agricultural land. Open in new tab To test for the program effect on allocative efficiency, I construct household level TFP for cultivating households in the sample using reported data on total farm area cultivated, farm output, and input expenditures. I first assume inputs can be converted into output through the following Cobb–Douglas function: $$\begin{eqnarray} \log {{Output}}_{i}=\theta _{l} \log l_{i}+\theta _{k} \log k_{i}+\theta _{h} \log h_{i}. \end{eqnarray}$$ Land l is the number of acres of operational farm size (whether owned or otherwise). Labor h is the sum of hired and family labor in number of days. Hired labor is the household’s expenditures on farm workers divided by the median daily wage in the district for any survey year. I count total number of days worked by household members who report being unpaid family laborers with primary occupations in farming to obtain the measure of family labor in days. Since family members only report days worked in the last month, I use the median number of months worked by agricultural farm workers to arrive at the total family labor days in the previous year. Capital k is the sum of the value of rented capital and expenses, adjusting for owned capital (expenses include intermediate inputs, like seeds, pesticides, and fertilizer). I do not have a measure of value of owned farmed machinery, so I use a proxy ψ calculated by regressing the log of total observed farm output on indicators for ownership of various assets.16 I then assume that the values of expenses and owned capital are separable and have the same factor share in production. To compute log Output, I choose θk = 0.11, θl = 0.25, and θh = 0.31 from the input elasticities for capital (k), land (l), and labor (h) calculated in Shenoy (2017) for rice farmers in Thailand.17 Weather variability and farmer-specific shocks are expected to be important factors contributing to realized output. To account for weather shocks common to all farmers in the same village, I regress the value of the difference between logged realized output Yi and computed log Outputi from the above production function on village-year fixed effects. I use the residual from this regression as my measure of farm TFP (in logs). Figure 3 shows the distribution of log TFP from 2011. Given the individual TFPs for each farming household, I categorize households by quartiles of the TFP distribution in a specific district and year. I later use this ranking to test if land is allocated toward farmers in higher TFP quartiles.18 Figure 3. Open in new tabDownload slide Distribution of farmer TFPs. Using the reported output Yi and operational land input, the marginal product of land for each farmer is calculated as follows: |${{MPL}}_{i}=\theta _{l}{Y_{i}}/{l_{i}}$|⁠. Agricultural Production I obtain aggregate crop output data from the Agricultural Statistics of Pakistan, which record the overall production and area cultivated for each crop at the district and year level. I also obtain normalized difference vegetation index (NDVI), a disaggregated measure of greenness, for all the study regions and over the study time period. These remote sensing data are based on 24 images per year from an average of 29,000 pixels per subdistrict, distributed by NASA (Didan 2015). I construct a subdistrict by year measure of the NDVI by using the following steps: I first get an average NDVI across all pixels within a subdistrict for each point the images are taken. Since the images are taken every 16 days, there are 24 such images per pixel in each year, or averaged NDVI for each subdistrict is available at 24 different points in any year. I allow for differences in agricultural seasons across regions and over time, and take the maximum NDVI across the 24 measurements over the year to obtain my measure of NDVI for each subdistrict and year. The NDVI measure is standardized to have mean 0 and standard deviation 1 across the subdistricts. Soil Quality For identification checks, I obtain remote sensing data for soil quality characteristics published by the International Institute for Applied Systems Analysis and part of the Harmonized World Soil Database. These evaluate soil quality according to the following criteria: nutrient availability, nutrient retention capacity, rooting conditions, oxygen availability to roots, excess salts, toxicity, and workability. For each of the seven dimensions, I compute the average value for all cells in each subdistrict’s perimeter and construct an index for each subdistrict using a principal components analysis. Administrative Data from Land Record Service Centers I use two sources of data from the Land Record Management and Information System database. First, I have visit level records from all service centers for the year 2016. These data include over 400,000 visits in the 12-month period, with a unique identity for each visitor, the center they patronized, and the nature of services received. In addition, I have a land parcel level list from the land records database for 18 of the 34 districts in the province. Primary Data on Farmers I use data from a phone survey with a subsample of over a million farmers from Punjab province comprising farm households enrolled in various government programs for which they submitted contact numbers. In mid-2020, phone surveys from roughly 1,800 randomly selected farmers solicited information on access to titles and land record service centers in their localities. These data identify landowners, whether they have a title for their property, and when they obtained the title. I construct a retrospective panel of households’ title ownership for households who were untitled before the program to test if the timing of when they obtain the title is correlated with the timing of the program in their district. I construct a dummy “Titled” at the household year level for ten years spanning the program rollout in Punjab. “Titled” takes on value 0 for untitled households until the year they obtain a title, after which it takes on value 1. 3.2. Theoretical Predictions Online Appendix C illustrates a simple framework to provide the intuition for the effects of improving tenure security on land allocation, based on Restuccia and Santaeulalia-Llopis (2017). Market frictions are incorporated as transaction costs in land leasing, and the framework predicts that under high transaction costs, low productivity farmers operate larger than optimal land as they are restricted from renting out, while productive farmers operate smaller than optimal land and have high marginal product of land. Thus, transaction costs manifest as wedges in the marginal products across farmers. If transaction costs were infinite, everyone would cultivate their land endowment. The reform is expected to lower transaction costs and improve market participation of agricultural households. In particular, high TFP (high MPL) households can have access to more land through the market, while low productivity farmers rent out land and reduce participation in agriculture. As a result of land mobility, farm scale and input choices may respond. Increasing market activity moves the allocation closer to the efficient allocation and lowers the wedges in the marginal products, which implies that aggregate level dispersion in marginal products goes down. Improved allocation of land and improved selection in farming both imply that aggregate production is higher. Farm-level output and yield may be affected through two channels. On one hand, the selection effect means more productive (high TFP) farmers are farming, while the least productive exit, implying higher farm output and yield for those who stay in farming. On the other hand, if there is an inverse farm size–yield relationship, land reallocation implies that farmers cultivate a larger amount of land on average and may a experience lower farm yield. Thus, the effect on average farm level yield is ambiguous, and may also differ across farmers. 3.3. Empirical Strategy I exploit the staggered rollout of the program by using a difference-in-difference (DD) strategy to compare trends in districts that received the program earlier relative to those that received it later. The program proposed one service center for each “tehsil” or subdistrict, but due to the nature of the data used, the subsequent analysis is at the district-year level. I use the fraction of subdistricts in a district d that have a functioning service center by year t to obtain program intensity at district level and run the following household level specification: $$\begin{eqnarray} \begin{array}{l}y_{{idt}}=\beta _{0}+\beta _{1}{{ProgramIntensity}}_{dt}+ X^{\prime }_{{idt}} \Psi +\mu _{d}+\eta _{t}+\mu _{d} \times t+\epsilon _{{idt}}, \end{array} \end{eqnarray}$$(1) where ProgramIntensity|$_{dt}$| is the percentage of subdistricts in a district with an active service center, and y|$_{idt}$| is an outcome for household i in district d and year t. X|$_{idt}$| are household demographic controls, and μd and ηt are district and year fixed effects, respectively. Household level controls include household head age, age-squared, education, and gender. To control for district-specific trends, I include an interaction of district fixed effects with a linear yearly trend. Standard errors are clustered at the district level. To account for the number of clusters, I also present wild-bootstrapped p-values (Cameron and Miller 2015). ProgramIntensity|$_{dt}$| = 1 indicates that all subdistricts in district d have the program. The coefficient β1 thus estimates an average treatment effect and represents the change in the outcome (beyond the district-specific trend and aggregate year fixed effects) due to an increase in program intensity. Identification is achieved from the variation in timing of program start as well as variation in the degree of program completion once it starts in any district. When subdistrict level data are available, I run a standard timing DD, where the primary independent variable is an indicator for the program at the subdistrict level. Identifying Assumptions The identification assumes that the timing of program start is quasi random; in particular, it is uncorrelated with district-specific trends after accounting for district and year fixed effects. I test for the validity of this identifying assumption with two different balance tests. First, I test for balance in pre-program characteristics across the various timing groups. Specifically, I use data from the pre-2011 survey round to regress the districtlevel outcomes of interest on fixed effects for each start year group. Standard errors are clustered at the district level. Columns (1)–(2) of Online Appendix Table A.2 panel A show the F-statistics from a test of the joint significance of the start year group fixed effects and the corresponding p-values. These regressions test if the timing of program start is correlated with the prior levels of the main outcomes and macro-economic variables. Second, I test if the timing of program start is correlated with prior changes in these outcomes. To test the relationship between program timing and prior trends, I regress the district level change in the main outcomes of interest on indicators for the year of program start. Columns (3)–(4) of Online Appendix Table A.2 panel A show the F-statistics from a test of the joint significance of the start year group fixed effects and the corresponding p-values. In addition to the district level outcomes, two remote-sensing variables (Soil Quality Index and NDVI) are available at the subdistrict level. I thus conduct the balance test for these outcomes with respect to program start at the subdistrict level. For all different outcomes, we can reject that the program start year fixed effects are jointly significant. Thus, the timing of treatment across districts does not appear to be driven by the level or changes in the main outcomes of interest, key macro-economic variables or underlying soil quality and productivity. In addition to the variation in timing, the identification strategy uses variation in the number of centers opened as a share of total subdistricts in a district. The identification thus assumes that the opening of centers at the subdistrict level is quasi-random. In the absence of subdistrict level data, I test this assumption by conducting balance tests for program intensity similar to above. I regress |${{ProgramIntensity}}$| on the prior year levels and changes of district level outcomes, district and year fixed effects.19 The results of these tests are presented in panel B of Online Appendix Table A.2. The balance tests confirm the validity of this assumption, as changes in |${{ProgramIntensity}}$| are not driven by the prior levels or changes in main outcomes of interest as the coefficients on various district level outcomes are small and mostly statistically insignificant.20 In similar spirit, I construct a placebo treatment variable by assuming program rollout prior to the actual launch of the program. If the differences in outcomes that are correlated with program intensity were pre-existing, we would see a significant correlation between the outcomes and the placebo program intensity variable. I later confirm that the program effects are unlikely to be spurious as the placebo program variable has no significant effects. I further account for factors that may compound the treatment effect by controlling for district-specific linear trends in all the regressions. There may be some concerns about district specific macro-economic cycles, as the study time-period represents a period of recovery from the 2008 global recession. In additional robustness checks, I control for macro-economic variables at the district level, quadratic district-level trends, and account for district-specific economic recovery by allowing for pre- and post-recession district-specific linear trends. I also conduct a placebo test with urban households to alleviate concerns that the program effects are capturing macro-economic trends across districts. These robustness checks are discussed in additional detail in Section 5 and the findings from the preferred specification are robust to these checks. Additionally, I estimate an event study specification to explicitly test for pre-existing differential trends and identify a post-program shift in trend. Last, I employ two additional identification strategies to compliment the findings from the main empirical analysis. First, I present the findings from the standard timing DD strategy, replacing ProgramIntensity|$_{dt}$| in equation (1) with a dummy PostProgram|$_{dt}$| that switches from 0 to 1 when the program starts in district d.21 Second, I use an alternate stacked DD identification strategy following Deshpande and Li (2019). The main findings are confirmed by the results of the complementary identification strategies. To account for multiple hypotheses being tested, I adjust my p-values following Anderson (2008) and Benjamini and Hochberg (1995). The details of the additional empirical strategies and tests are presented in Section 5. For land sales and rental market participation, I limit the regression sample to landowning households, while for farm input and output (farm size, crop and input choices, and yield), the sample includes cultivating households (including both landowners and landless cultivators). An additional concern arises if these samples are shifting over time, particularly in response to the program. I test the program effect on an indicator for inclusion in the rural, ‘landowners’ and ‘cultivators’ sample in Online Appendix Table A.3; these tests provide assurance that the likelihood of being a rural, or a landowner or farming household among rural households does not respond to the program. 4. Results 4.1. Take-Up of the Program The argued mechanisms for the program effects are improved access to land records and tenure security as a result of land record digitization. In this section, I support these proposed “first-stage” effects with both qualitative and empirical evidence on access and usage of the land record service centers. First, landowner reported data prior to the program and qualitative reports from patwaris indicate that access to titles and property rights security was the dominant constraint resolved by the program. After the rollout of the reform, 72% of visitors to land record service centers perceive the new system to be more secure, and 60% also report service centers are expected to reduce land disputes.22 Qualitative interviews of land revenue officers including patwaris also stand testament to the increased access to land titles and rights verification due to the program. These observations suggest that the reform improved tenure security by obviating manual manipulation of records, diminishing the role of patwaris and other revenue officers, and facilitating improved access and verifiability of ownership and cultivation rights. Second, I observe substantial usage of the service centers shortly after they are operational. The service centers received over 400,000 unique visitors in 2016, with an average ratio of visitors to land parcels of 18% over the 12 month period. I note that the same household or landowner can own multiple parcels so the ratio of visits to landowners is expected to be higher. The type of services obtained at the service centers show that 70% of all visits are for the purpose of obtaining a title, and another 10% are for confirmation of land rights. Last, I document that service centers led to an increase in title ownership. In the phone survey sample, 75% of the landowners have some title or “fard” for their property in 2020—56% have a computerized title from the land record center, while the rest have the old or “manual” version of the title. Based on farmer responses, I infer that before computerized records were available, 46% of farmers did not a have a title for their land. Of the untitled landowners, about 45% had obtained a title by the time of the survey, a substantial increase in title ownership over the 4 to 8 year period for which the program had been operating across different regions. For the causal effect of the land records program on access to land titles, I regress “Titled” on “Post” or an indicator that switches to 1 when the program starts in the subdistrict of the landowning households’ agricultural property, household fixed effects, and year fixed effects interacted with household’s landholding. Consistent with the other analyses, I cluster standard errors at the district level and wild-bootstrap the standard errors. This clustering also accounts for the serial correlation in the outcome within households. The findings from this regression, shown in Online Appendix Table A.4, show that the opening of the service center is associated with a 2 percentage point increase in acquiring a title. This amounts to 16% of the title ownership rate among the sample households by the end of the regression period. This effect may be an underestimate as we expect title ownership to continue to increase after the rollout of the program, whereas the DD specification estimates the increase in title access in the year the program dummy switched from 0 to 1. Self-reported “take-up” is also high. In the entire phone survey sample (including some farmers who do not own land), 50% have used the land record service center and 75% of these respondents patronized the service centers for obtaining a title or record confirmation.23 “Record confirmation” is typically reported by cultivators who are not land owners, which suggests that property rights are more accessible and transparent for tenants as well. These data together provide an underpinning for the expected effects on tenure security and land market activity. 4.2. Program Effects on Land and Labor Market Participation by Landowners Lack of ownership security restricts landowners from trading their land, that is, renting out or selling it. Only 21% of landowning households report renting out their land, while only about 1% report having sold or purchased a portion of their agricultural land holding in the prior year. Among cultivating households, 15% are landless, and a third report renting in any land for cultivation. To examine the land market effects of the program, I first test if land ownership shifts as the program is rolled out. The first outcome in Table 2 is an indicator for land ownership among all rural households. The program has no effect on the rate of land ownership, which could be consistent with no market activity or market transactions that caused land ownership to shift across households without changing the overall rate of ownership. To investigate this, I consider the change in recent land transactions by landowning households as the program is rolled out. I find that the rate of land purchase (column 2) or sale (column 3) does not respond to the program, suggesting that the program does not change the constraints on land ownership transactions.24 Table 2. Program effect on market activity for land owners. . Own agricultural land . Agricultural land purchased . Agricultural land sold . Agricultural land rentout . . (Y/N) . (Y/N) . (Y/N) . (Y/N) . . (1) . (2) . (3) . (4) . Program intensity 0.002 0.001 −0.002 0.061** (0.030) (0.003) (0.006) (0.027) [0.954] [0.821] [0.707] [0.0327] Observations 19,067 7,584 7,584 7,597 Mean dep., pre-program 0.420 0.006 0.010 0.219 Sample households All rural All landowning All landowning All landowning . Own agricultural land . Agricultural land purchased . Agricultural land sold . Agricultural land rentout . . (Y/N) . (Y/N) . (Y/N) . (Y/N) . . (1) . (2) . (3) . (4) . Program intensity 0.002 0.001 −0.002 0.061** (0.030) (0.003) (0.006) (0.027) [0.954] [0.821] [0.707] [0.0327] Observations 19,067 7,584 7,584 7,597 Mean dep., pre-program 0.420 0.006 0.010 0.219 Sample households All rural All landowning All landowning All landowning Notes: All regressions include district and year fixed effects, and controls for linear district-level yearly trends. Additional household controls include head age, age squared, education, and gender. Data are from the HIES surveys. Standard errors clustered at the district level in parentheses. Wild cluster bootstrapped p-values in brackets. **p < 0.05. Open in new tab Table 2. Program effect on market activity for land owners. . Own agricultural land . Agricultural land purchased . Agricultural land sold . Agricultural land rentout . . (Y/N) . (Y/N) . (Y/N) . (Y/N) . . (1) . (2) . (3) . (4) . Program intensity 0.002 0.001 −0.002 0.061** (0.030) (0.003) (0.006) (0.027) [0.954] [0.821] [0.707] [0.0327] Observations 19,067 7,584 7,584 7,597 Mean dep., pre-program 0.420 0.006 0.010 0.219 Sample households All rural All landowning All landowning All landowning . Own agricultural land . Agricultural land purchased . Agricultural land sold . Agricultural land rentout . . (Y/N) . (Y/N) . (Y/N) . (Y/N) . . (1) . (2) . (3) . (4) . Program intensity 0.002 0.001 −0.002 0.061** (0.030) (0.003) (0.006) (0.027) [0.954] [0.821] [0.707] [0.0327] Observations 19,067 7,584 7,584 7,597 Mean dep., pre-program 0.420 0.006 0.010 0.219 Sample households All rural All landowning All landowning All landowning Notes: All regressions include district and year fixed effects, and controls for linear district-level yearly trends. Additional household controls include head age, age squared, education, and gender. Data are from the HIES surveys. Standard errors clustered at the district level in parentheses. Wild cluster bootstrapped p-values in brackets. **p < 0.05. Open in new tab Improving tenure security can increase the likelihood of tenancy transactions even if land ownership stays stable. Column (4) in Table 2 shows that among landowning households, the likelihood of renting out increases by 6 percentage points when the program is completed in their district. This is a large effect, given the 22% rate of renting out on average across the districts prior to the program. Landowners renting out could be those who previously owned land or households that are able to purchase more land due to the program and then rent it out. Since there are no significant effects on the agricultural land ownership rate, ownership transactions, or the average size of land-owned, we can deduce that the change in tenancy is driven by previous landowners. Thus, the program resolved land market frictions that constrained existing landowners from renting out. Relieving constraints on renting out for existing landowners can have spillover effects on the labor market. Specifically, agricultural participation is allegedly high due to insecure property rights on agricultural land that prevent households from participating in off-farm activities for better income, as vacating land bears the risk of losing it (Field 2007). Increased rental activity by landowning households implies some landowning household members no longer need to practice cultivation if they have opportunities for participating in non-farm activities. The next set of results in Table 3 examines the effect of the program on participation in agricultural activities by landowning households. Consistent with high likelihood of renting out, I find that on average, these households are less likely to participate in agriculture as a result of the program. Three different outcome variables indicate this. Households are less likely to cultivate a farm (column 1), less likely to “self-cultivate” or cultivate owned land (column 2), and less likely to to have members that participate in agriculture broadly, including wage work (column 3). In sum, landowners are, on average, 27% more likely to rent out their agricultural land and 12% more likely to quit agriculture due to the program. Column 4 of Table 3 shows the intensive margin measured by the share of households’ income from agricultural activities. Consistent with the changes in the occupational choices of landowning households, the proportion of income from agricultural activities falls by 7 percentage points, when comparing wholly treated districts to wholly untreated ones. This corresponds to a 12% drop in income share of landowning households from agriculture. Table 3. Program effect on agricultural participation. . Household operates any farm . Household operates owned land . Household member agricultural worker . Share income from agriculture . . (1) . (2) . (3) . (4) . Program intensity −0.098*** −0.089*** −0.099*** −0.080** (0.030) (0.032) (0.035) (0.034) [0.000900] [0.00660] [0.0114] [0.0315] Observations 7,597 7,597 7,597 7,597 Mean dep., pre-program 0.786 0.756 0.807 0.650 . Household operates any farm . Household operates owned land . Household member agricultural worker . Share income from agriculture . . (1) . (2) . (3) . (4) . Program intensity −0.098*** −0.089*** −0.099*** −0.080** (0.030) (0.032) (0.035) (0.034) [0.000900] [0.00660] [0.0114] [0.0315] Observations 7,597 7,597 7,597 7,597 Mean dep., pre-program 0.786 0.756 0.807 0.650 Notes: Sample includes all agricultural landowning households. All regressions include district and year fixed effects, and controls for linear district-level yearly trends. Additional household controls include head age, age squared, education, and gender. Standard errors clustered at the district level in parentheses. Wild cluster bootstrapped p-values in brackets. ***p < 0.01, **p < 0.05. Open in new tab Table 3. Program effect on agricultural participation. . Household operates any farm . Household operates owned land . Household member agricultural worker . Share income from agriculture . . (1) . (2) . (3) . (4) . Program intensity −0.098*** −0.089*** −0.099*** −0.080** (0.030) (0.032) (0.035) (0.034) [0.000900] [0.00660] [0.0114] [0.0315] Observations 7,597 7,597 7,597 7,597 Mean dep., pre-program 0.786 0.756 0.807 0.650 . Household operates any farm . Household operates owned land . Household member agricultural worker . Share income from agriculture . . (1) . (2) . (3) . (4) . Program intensity −0.098*** −0.089*** −0.099*** −0.080** (0.030) (0.032) (0.035) (0.034) [0.000900] [0.00660] [0.0114] [0.0315] Observations 7,597 7,597 7,597 7,597 Mean dep., pre-program 0.786 0.756 0.807 0.650 Notes: Sample includes all agricultural landowning households. All regressions include district and year fixed effects, and controls for linear district-level yearly trends. Additional household controls include head age, age squared, education, and gender. Standard errors clustered at the district level in parentheses. Wild cluster bootstrapped p-values in brackets. ***p < 0.01, **p < 0.05. Open in new tab I also test the changes in the alternate occupational choices of landowning households as they are able to rent out their agricultural land and exit agriculture. Online Appendix Table A.5 shows an increase in the share of household members that participate in non-agricultural activities. These are statistically significant at the 10% level for participation in large business ownership, and positive (but statistically insignificant) for participation in small businesses, self-employment or as paid employees.25 In summary, the results above demonstrate that weak property rights constrain landowners from leaving agriculture, forcing them to cultivate their owned land instead of renting out land and engaging in other economic activities. Improved ownership security through the computerization of ownership and tenancy rights reduced market frictions, allowing landowners to rent out their land and increase participation in non-agricultural activities. I test the heterogenous effects of the program by income quartile in the Online Appendix Tables A.7 and A.8. Heterogeneous effects demonstrate that land market frictions are particularly extreme for poorer, and plausibly less-connected, households. The program improves land rental probability and reduces agricultural participation for households in the lowest income quartile; the richest households experience lower impact on both land rental and labor participation, relative to the poorest households. These effects are consistent with the motivation behind the design of this computerization program, which intended to increase accessibility of records for the marginalized sections of the rural population. 4.3. Program Effects on Farm Operation by Cultivators The next set of regressions estimate the program effects for cultivating households, which includes landowning households that stay in cultivation as well as landless farm households. Table 4 shows the effect of the program on the intensive margin of renting in, measured by average quantity of rented in land among cultivating households. The program has a strong positive effect on land rented in on fixed cash rent, and no significant effect on land that is sharecropped. This is consistent with the view that land owners with less secure property rights may choose sharecropping, as it allows landlords to exert stricter property control by bearing a higher amount of production risk than in fixed rent contracts (Bellemare 2012). Sharecropping is also typically arranged between landlords and tenants in the same village due to the sharing nature of this tenancy arrangement and for ease of monitoring; thus, the threat of weak property rights might be less binding for sharecropping.26 Table 4. Program effect on farm size and rented in land. . Rented . Sharecropped . Owned . Total cultivated . . (1) . (2) . (3) . (4) . Program intensity 0.925** 0.084 0.731 1.110** (0.433) (0.255) (0.697) (0.452) [0.0351] [0.797] [0.320] [0.0151] Observations 7,256 7,256 7,256 7,256 Mean dep., pre-program 1.648 0.686 5.423 7.055 . Rented . Sharecropped . Owned . Total cultivated . . (1) . (2) . (3) . (4) . Program intensity 0.925** 0.084 0.731 1.110** (0.433) (0.255) (0.697) (0.452) [0.0351] [0.797] [0.320] [0.0151] Observations 7,256 7,256 7,256 7,256 Mean dep., pre-program 1.648 0.686 5.423 7.055 Notes: Sample includes all cultivating households. Rented corresponds to area under fixed cash rent contracts and sharecropped refers to area under sharecropping contracts. Total cultivated is total operational farm area including owned land. All regressions include district and year fixed effects, and controls for linear district-level yearly trends. Additional household controls include head age, age squared, education, and gender. Data are from the HIES surveys. Standard errors clustered at the district level in parentheses. Wild cluster bootstrapped p-values in brackets. **p < 0.05. Open in new tab Table 4. Program effect on farm size and rented in land. . Rented . Sharecropped . Owned . Total cultivated . . (1) . (2) . (3) . (4) . Program intensity 0.925** 0.084 0.731 1.110** (0.433) (0.255) (0.697) (0.452) [0.0351] [0.797] [0.320] [0.0151] Observations 7,256 7,256 7,256 7,256 Mean dep., pre-program 1.648 0.686 5.423 7.055 . Rented . Sharecropped . Owned . Total cultivated . . (1) . (2) . (3) . (4) . Program intensity 0.925** 0.084 0.731 1.110** (0.433) (0.255) (0.697) (0.452) [0.0351] [0.797] [0.320] [0.0151] Observations 7,256 7,256 7,256 7,256 Mean dep., pre-program 1.648 0.686 5.423 7.055 Notes: Sample includes all cultivating households. Rented corresponds to area under fixed cash rent contracts and sharecropped refers to area under sharecropping contracts. Total cultivated is total operational farm area including owned land. All regressions include district and year fixed effects, and controls for linear district-level yearly trends. Additional household controls include head age, age squared, education, and gender. Data are from the HIES surveys. Standard errors clustered at the district level in parentheses. Wild cluster bootstrapped p-values in brackets. **p < 0.05. Open in new tab The program proves to relieve the constraints in the fixed rent lease market for agricultural land. Column (3) shows that owned cultivated area has no statistically significant change, which is consistent with the absence of any effects on land sales. Finally, column (4) suggests that as more land is rented in, average farm size increases, indicating meaningful impacts of the program on scale of agriculture in Punjab. Average operational farm size is about 1 acre or 15% higher just after the program’s completion in a district. In the Online Appendix, I explore heterogeneity across households in these outcomes; Online Appendix Table A.11 presents these effects and shows that among cultivating households, landless households benefit from greater access to land due to improved rental markets (marginally more than landed farm households). 4.4. Program Effect on Land Allocation across Farmers To test the program effect on the allocation of land across farmers, I re-run specification (1) interacting program intensity with TFP quartiles: $$\begin{eqnarray} y_{{idt}} &= & \phi _{0}+\phi _{1,j}{{ProgramIntensity}}_{{dt}} \times {{TFP Quartile}}_{{ij}} +\phi _{2,j} {{TFP Quartile}}_{{ij}} \nonumber\\ & & + X^{\prime }_{{idt}} \Psi +\mu _{d}+\eta _{t}+\mu _{d} \times t+\epsilon _{{idt}}, \end{eqnarray}$$(2) y measures total cultivated land, or MPL for any farmer. TFPQuartileij is an indicator for a farmer i being in any quartile j of the TFP distribution in a district-year cell.27 The theoretical framework predicts that reducing transaction costs in the leasing market will induce lower TFP farmers to rent out land and exit agriculture, while higher TFP farmers have greater access to land and lower marginal product of land. Thus, in equation (2), we expect φ1 to be positive for land and negative for MPL for farmers in higher TFP quartiles.28 Table 5 shows the effect of the program on land allocation across farmers at different parts of the productivity distribution. The table shows the log of total cultivated area (or farm size in acres) and the marginal products of land as a function of program intensity interacted with each farmers’ TFP quartile. The findings in Table 5 indicate that farmers in the higher TFP quartiles are more likely to have greater access to land, as indicated by high operational farm size and lower marginal product of land. In the highest TFP quartile, farm size increases by 24% with the program. Table 5. Program effect on allocation across farmers. . Land . MPL . . (1) . (2) . Program intensity 0.050 0.154 (0.089) (0.099) [0.579] [0.134] TFP quartile 2 × program intensity 0.081 −0.081* (0.074) (0.044) [0.273] [0.0716] TFP quartile 3 × program intensity 0.096 −0.130** (0.064) (0.054) [0.139] [0.0202] TFP quartile 4 × program intensity 0.244** −0.244*** (0.102) (0.059) [0.0281] [0.000200] Observations 7,256 7,256 . Land . MPL . . (1) . (2) . Program intensity 0.050 0.154 (0.089) (0.099) [0.579] [0.134] TFP quartile 2 × program intensity 0.081 −0.081* (0.074) (0.044) [0.273] [0.0716] TFP quartile 3 × program intensity 0.096 −0.130** (0.064) (0.054) [0.139] [0.0202] TFP quartile 4 × program intensity 0.244** −0.244*** (0.102) (0.059) [0.0281] [0.000200] Observations 7,256 7,256 Notes: Sample includes all cultivating households. All regressions include district and year fixed effects, and controls for linear district-level yearly trends. Additional household controls include head age, age squared, education, and gender. Data are from the HIES surveys. Standard errors clustered at the district level in parentheses. Wild cluster bootstrapped p-values in brackets. Farmer TFP is calculated as demonstrated in Section 3.1, and TFP quartiles are calculated within the district. *p < 0.1, **p < 0.05, ***p < 0.01. Open in new tab Table 5. Program effect on allocation across farmers. . Land . MPL . . (1) . (2) . Program intensity 0.050 0.154 (0.089) (0.099) [0.579] [0.134] TFP quartile 2 × program intensity 0.081 −0.081* (0.074) (0.044) [0.273] [0.0716] TFP quartile 3 × program intensity 0.096 −0.130** (0.064) (0.054) [0.139] [0.0202] TFP quartile 4 × program intensity 0.244** −0.244*** (0.102) (0.059) [0.0281] [0.000200] Observations 7,256 7,256 . Land . MPL . . (1) . (2) . Program intensity 0.050 0.154 (0.089) (0.099) [0.579] [0.134] TFP quartile 2 × program intensity 0.081 −0.081* (0.074) (0.044) [0.273] [0.0716] TFP quartile 3 × program intensity 0.096 −0.130** (0.064) (0.054) [0.139] [0.0202] TFP quartile 4 × program intensity 0.244** −0.244*** (0.102) (0.059) [0.0281] [0.000200] Observations 7,256 7,256 Notes: Sample includes all cultivating households. All regressions include district and year fixed effects, and controls for linear district-level yearly trends. Additional household controls include head age, age squared, education, and gender. Data are from the HIES surveys. Standard errors clustered at the district level in parentheses. Wild cluster bootstrapped p-values in brackets. Farmer TFP is calculated as demonstrated in Section 3.1, and TFP quartiles are calculated within the district. *p < 0.1, **p < 0.05, ***p < 0.01. Open in new tab A more efficient allocation implies that at a market level, the dispersion of marginal products would go down.29 In order to test this prediction, I construct measures of MPL dispersion by calculating the standard deviation, coefficient of variation, and interquartile range within a district in each survey year. The sample size restrictions do not allow the construction of these measures at a village level. (Rental market operation may be more likely to occur at the village level, but the sample size within the same village is too small.) In Table 6, I find a statistically significant negative effect on MPL dispersion measured in three different ways. Additional effect on the dispersion of TFPs is also negative with moderate level of significance, and shown in Online Appendix Table A.12. These outcomes are consistent with Table 5, showing a reallocation of land across farmers, and also corroborate the earlier evidence on reallocation of some households away from farming. Table 6. Dispersion of marginal product of land. . SD . CV . 75−25 . . (1) . (2) . (2) . Program intensity −0.134** −0.013* −0.110 (0.065) (0.007) (0.072) [0.0112] [0.0133] [0.0525] Observations 170 170 170 . SD . CV . 75−25 . . (1) . (2) . (2) . Program intensity −0.134** −0.013* −0.110 (0.065) (0.007) (0.072) [0.0112] [0.0133] [0.0525] Observations 170 170 170 Notes: All regressions include district and year fixed effects, and district-level linear trend. District-level controls include average education and rate of land ownership. Data are from the HIES surveys. Standard errors clustered at the district level in parentheses. Wild cluster bootstrapped p-values in brackets. *p < 0.1, **p < 0.05. Open in new tab Table 6. Dispersion of marginal product of land. . SD . CV . 75−25 . . (1) . (2) . (2) . Program intensity −0.134** −0.013* −0.110 (0.065) (0.007) (0.072) [0.0112] [0.0133] [0.0525] Observations 170 170 170 . SD . CV . 75−25 . . (1) . (2) . (2) . Program intensity −0.134** −0.013* −0.110 (0.065) (0.007) (0.072) [0.0112] [0.0133] [0.0525] Observations 170 170 170 Notes: All regressions include district and year fixed effects, and district-level linear trend. District-level controls include average education and rate of land ownership. Data are from the HIES surveys. Standard errors clustered at the district level in parentheses. Wild cluster bootstrapped p-values in brackets. *p < 0.1, **p < 0.05. Open in new tab Since the construction of TFP requires a number of assumptions, I examine land allocation in response to the program using two alternate measures of productivity: agricultural yield per acre and profits per acre. These results are in Online Appendix Table A.13 and show that land is higher for all households that stay in farming after the program, but significantly higher for farmers in the 3rd and 4th quartiles of the productivity distribution, whereas MPL is lower for them. These results are consistent with the program effects estimated across farmers ranked by TFP. Together, the findings provide credible support for the hypothesis that an improved allocation of land across farmers is underlying the average effects on rental market and farm size. 4.5. Program Effect on Farm-Level and Aggregate Output and Yield The effect of the program on farming scale has important implications for agricultural input choices, mechanization, and productivity improvement, as farm size is assumed to be a constraint to adoption of capital intensive technologies (Foster and Rosenzweig 2011). Moreover, insecurity of tenure affects investment incentives, as Jacoby and Mansuri (2008) demonstrate that non-contractible investments are under-provided on leased land in Pakistan due to incomplete contracts. If optimal farm area and/or increased tenure security induces higher input, especially capital usage on farms, then output should increase. Even if the capital margin is unaffected, improved allocative efficiency will result in higher aggregate productivity. I examine the effects of the program on farm output in Table 7 and aggregate output in Tables 8 and 9. The farm output, yield, and profits are expected to be measured with error, and winsorized values are used to test for program effects. Table 7. Program effect on farm-level agricultural production. . Total output . Output per acre . Profit per acre . . (1) . (2) . (3) . Program intensity 90.439*** 3.216 3.906 (32.719) (5.270) (4.249) [0.0109] [0.550] [0.378] Observations 7,256 7,256 7,256 Mean dep., pre-program 156.338 25.611 15.514 . Total output . Output per acre . Profit per acre . . (1) . (2) . (3) . Program intensity 90.439*** 3.216 3.906 (32.719) (5.270) (4.249) [0.0109] [0.550] [0.378] Observations 7,256 7,256 7,256 Mean dep., pre-program 156.338 25.611 15.514 Notes: Output for each farm is calculated as the sum of the value of all crops grown on the farm. The value of each crop is calculated using the yield times median price for the crop across all farm households. All regressions include district and year fixed effects, and controls for linear district-level yearly trends. Additional household controls include head age, age squared, education, and gender. Data are from the HIES surveys. Standard errors clustered at the district level in parentheses. Wild cluster bootstrapped p-values in brackets. ***p < 0.01. Open in new tab Table 7. Program effect on farm-level agricultural production. . Total output . Output per acre . Profit per acre . . (1) . (2) . (3) . Program intensity 90.439*** 3.216 3.906 (32.719) (5.270) (4.249) [0.0109] [0.550] [0.378] Observations 7,256 7,256 7,256 Mean dep., pre-program 156.338 25.611 15.514 . Total output . Output per acre . Profit per acre . . (1) . (2) . (3) . Program intensity 90.439*** 3.216 3.906 (32.719) (5.270) (4.249) [0.0109] [0.550] [0.378] Observations 7,256 7,256 7,256 Mean dep., pre-program 156.338 25.611 15.514 Notes: Output for each farm is calculated as the sum of the value of all crops grown on the farm. The value of each crop is calculated using the yield times median price for the crop across all farm households. All regressions include district and year fixed effects, and controls for linear district-level yearly trends. Additional household controls include head age, age squared, education, and gender. Data are from the HIES surveys. Standard errors clustered at the district level in parentheses. Wild cluster bootstrapped p-values in brackets. ***p < 0.01. Open in new tab Table 8. Program effect on aggregate agricultural production using remote sensing NDVI. . (1) . (2) . Post Program 0.093** 0.093** (0.040) (0.041) Observations 1,792 1,792 Linear trend District Sub-district . (1) . (2) . Post Program 0.093** 0.093** (0.040) (0.041) Observations 1,792 1,792 Linear trend District Sub-district Notes: Regressions are at subdistrict-year level, and the outcome is the NDVI, measured in number of standard deviations. All regressions include subdistrict and year fixed effects, in addition to the linear trends mentioned in the table. PostProgram is an indicator for all years after the program starts in any subdistrict. NDVI data are from (Didan 2015). Standard errors clustered at the subdistrict level are presented in parentheses. **p < 0.05. Open in new tab Table 8. Program effect on aggregate agricultural production using remote sensing NDVI. . (1) . (2) . Post Program 0.093** 0.093** (0.040) (0.041) Observations 1,792 1,792 Linear trend District Sub-district . (1) . (2) . Post Program 0.093** 0.093** (0.040) (0.041) Observations 1,792 1,792 Linear trend District Sub-district Notes: Regressions are at subdistrict-year level, and the outcome is the NDVI, measured in number of standard deviations. All regressions include subdistrict and year fixed effects, in addition to the linear trends mentioned in the table. PostProgram is an indicator for all years after the program starts in any subdistrict. NDVI data are from (Didan 2015). Standard errors clustered at the subdistrict level are presented in parentheses. **p < 0.05. Open in new tab Table 9. Program effect on aggregate agricultural production using district-level data. . Cereal crops . Cash crops . . Log area . Log output . Log yield . Log area . Log output . Log yield . . (1) . (2) . (3) . (4) . (5) . (6) . Program intensity 0.009 0.061* 0.053* −0.023 −0.044 −0.021 (0.024) (0.031) (0.028) (0.093) (0.125) (0.042) [0.702] [0.0572] [0.0816] [0.824] [0.727] [0.646] Observations 792 792 792 455 455 455 . Cereal crops . Cash crops . . Log area . Log output . Log yield . Log area . Log output . Log yield . . (1) . (2) . (3) . (4) . (5) . (6) . Program intensity 0.009 0.061* 0.053* −0.023 −0.044 −0.021 (0.024) (0.031) (0.028) (0.093) (0.125) (0.042) [0.702] [0.0572] [0.0816] [0.824] [0.727] [0.646] Observations 792 792 792 455 455 455 Notes: Regressions are at the district-crop-year level, and the outcomes are logged total area, total output, and yield for each crop in each district and year. Cereal crops include maize, rice, and wheat, while cash crops include cotton and sugarcane. All regressions include district, crop and year fixed effects, and district-specific linear trends. District-level controls include average education and rate of landownership. Data are from the National Agricultural Statistics. Standard errors clustered at the district level in parentheses. Wild cluster bootstrapped p-values in brackets. *p < 0.1. Open in new tab Table 9. Program effect on aggregate agricultural production using district-level data. . Cereal crops . Cash crops . . Log area . Log output . Log yield . Log area . Log output . Log yield . . (1) . (2) . (3) . (4) . (5) . (6) . Program intensity 0.009 0.061* 0.053* −0.023 −0.044 −0.021 (0.024) (0.031) (0.028) (0.093) (0.125) (0.042) [0.702] [0.0572] [0.0816] [0.824] [0.727] [0.646] Observations 792 792 792 455 455 455 . Cereal crops . Cash crops . . Log area . Log output . Log yield . Log area . Log output . Log yield . . (1) . (2) . (3) . (4) . (5) . (6) . Program intensity 0.009 0.061* 0.053* −0.023 −0.044 −0.021 (0.024) (0.031) (0.028) (0.093) (0.125) (0.042) [0.702] [0.0572] [0.0816] [0.824] [0.727] [0.646] Observations 792 792 792 455 455 455 Notes: Regressions are at the district-crop-year level, and the outcomes are logged total area, total output, and yield for each crop in each district and year. Cereal crops include maize, rice, and wheat, while cash crops include cotton and sugarcane. All regressions include district, crop and year fixed effects, and district-specific linear trends. District-level controls include average education and rate of landownership. Data are from the National Agricultural Statistics. Standard errors clustered at the district level in parentheses. Wild cluster bootstrapped p-values in brackets. *p < 0.1. Open in new tab Table 7 shows positive effects on the total farm output, output per acre, and profits or value added, though the effect on output and profit per acre is not statistically significant.30 The increase in output is consistent with an increase in farm scale, while the null effect on farm-level yield and profit is consistent with the combined effect of a negative farm size productivity relationship (Foster and Rosenzweig 2017) and a selection of more productive farmers into farming. My measure of aggregate production, NDVI, is obtained from remote sensing data on vegetation, as explained in Section 3.1. Since these data are now at the subdistrict level, I run a subdistrict-year level regression, with an indicator PostProgram as the independent variable of interest, which is 1 for each year after the program has started in a subdistrict. I include subdistrict and year fixed effects and linear trends at the subdistrict and district level. The results, presented in Table 8, show the effect of the program on the NDVI in standard deviations, and show a robust positive effect on productivity as measured by the NDVI. Introducing the program increases the level of production at the subdistrict level by 9% of a standard deviation. Table 9 employs alternate, aggregate district-level crop production data from administrative sources. The regressions are at district-crop-year level, where the outcome of interest is log of crop yield (ton/ha) for each district by year and for the major cereal and cash crops31. The district-level yield regressions include crop, district, and year fixed effects as well as district-specific linear trends, and span all years from 2005 to 2015. I find that while the program does not affect total cultivated area under the major cereal or cash crops, aggregate cereal output and yield are 6% and 5% higher, respectively, when the program is completed in any district. These effects are significant at the 10% level. The major cash crops show an average 2% decline in yield, but this effect is statistically indistinguishable from 0. These district-level data provide supporting evidence of improved aggregate productivity, but suffer from caveats that are typical for government-collected administrative data. Online Appendix Tables A.14 and A.15 show the changes in inputs and crop choice underlying the farm size effects. Farmers shift into rice and away from planting maize on their land, as shown in Online Appendix Table A.14. Such a shift may or may not be expected, and there are not many consistent mechanisms that explain this shift. Second, the effect on input choices in Online Appendix Table A.15 shows an increase in the usage of pesticides (marginally statistically significant). The usage of rented equipment is lower, though not statistically significantly, which could suggest an increased likelihood that farmers’ use owned machinery and equipment. Since data on farmer ownership of agricultural machinery are not available, I cannot explicitly test this hypothesis. Farmers do report if they acquired (purchased or received) any agricultural machinery, including tube wells, tractors, ploughs, threshers, harvesters, or trucks, in the previous year. I test the effect of the program on acquisition of agricultural machinery and find that while there is no overall effect, landless households are particularly more likely to have acquired agricultural equipment when the program is completed in their districts (Online Appendix Table A.16). Altogether, greater farm size, higher usage of some inputs, increased owned equipment, and reduced equipment rental are consistent with increased incentives to invest in productive mechanized inputs. 4.6. Discussion of Findings and Mechanisms Evidence on take-up shows that the program’s effects are largely driven by an improvement of perceived property rights security through access to land records and titles. The survey data used for the analysis do not measure perceived property rights directly but ask landowners how much they expect to receive if they were to sell their land. I create a proxy for ability to sell one’s land using an indicator that equals 1 if the respondent reports some positive expected payment from the sale of their property.32 Using specification (1), I find a significant improvement in the proxy for perceived selling rights with the program. This is despite no change in land purchase/sale as shown earlier, or in the total area owned. The program has no impact on perceived rights for property types that were unaffected by the land records program, as measured by a similarly constructed proxy for non-agricultural land. This suggestive evidence (presented in Online Appendix Table A.17) in combination with other types of evidence provide a strong case for improved access to land rights and property rights security due to the program. The lack of effects on land ownership and sales can be due to a number of reasons. First, land sale and rental markets may be substitutes. Second, landowners may rent out their land, but not sell it, as lack of complete insurance and credit markets induce them to hold onto land as a precautionary asset (Rosenzweig 2001). Finally, the program effects are strongest for the lowest income quartiles among landowning households. These households may be less likely to participate in the sales market regardless of the program. If only large landowners participate in the land sales market, the sales margin is less likely to respond as large landowners are less likely to face tenure insecurity before the program and are therefore least affected by it. The results, put together, demonstrate that a light-touch reform has promising prospects for the outlook of agricultural land markets and structural change in Punjab. Aside from facilitating fixed cash rent transactions and increasing farming scale, the program can induce a productive shift in the allocation of labor, paving the way for improved agricultural productivity and transition of rural labor into non-agricultural sectors, which are necessary predecessors of transition in the economy. 5. Robustness and Alternate Identification Strategies I conduct a number of additional tests to validate the identification strategy and robustness of the findings. 5.1. Additional Time Periods I test the primary outcomes using an extended dataset that combines the HIES with the PSLM surveys. The outcomes that are measured in both surveys include land ownership, rental, and household occupation and are shown in Online Appendix Table A.18. I find that the likelihood of renting out and leaving agriculture are higher for households in the districts with the program and no change in land ownership, confirming the primary results. Further, in Online Appendix Table A.19 (panel A), I ensure that the choice of controls does not drive the main results by showing the effects from the main specification without household controls. 5.2. Alternate Macro-Economic Trends I address the concern that the study period coincides with a period of global recession followed by a recovery and that some of the program effects may be driven by differential rates of recovery across districts. I rule out differential business cycle events across districts in four ways. I allow for non-linear district-specific trends by controlling for quadratic trends by districts. I control directly for macro-economic outcomes at the district level in the main specification. Particularly, I add district-level unemployment and size of the labor force to the regressions. In an additional robustness check, I allow for pre- and post-recovery trends for districts. Online Appendix Figure A.2 shows the GDP per capita for Pakistan, which stagnates during global recession in 2007–2009 and recovers in the post-2010 period. I interact the district-specific trend with an indicator for the post recession period to allow for varying rates of recovery across districts. These results are presented in panels B-D of Online Appendix Table A.19 and demonstrate that the treatment effects of the program are not sensitive to these additional controls. The coefficient estimates are qualitatively and quantitively unmoved when I account for the possibly confounding macro-economic changes. Lastly, I conduct a placebo test using the income of urban households that is unlikely to be affected by the land records program and find a precisely estimated null effect (Online Appendix Table A.20). This provides further reassurance that the program rollout is not capturing a differential recovery from the global recession across districts. Similarly, the earliest program districts may have been selected endogenously as the pilot districts for a salient program and may have differential trends. I ensure that the treatment effects are robust to excluding the three districts where the program began in 2011 (Online Appendix Table A.21). 5.3. Event Study Analysis I test for pre-existing trends using an event study analysis using the expanded data set with the PSLM and HIES surveys that have data from consecutive years with the exception of 2009 when neither survey was conducted. Figure 4 shows the event study graphs, which plot the coefficients, γl, from the following district-level regression. $$\begin{eqnarray} y_{dt}=\gamma _{0}+\sum _{l}\gamma _{l}{{YearsSinceProgram}}_{dt,l}+\mu _{d}+\eta _{t}+\varepsilon _{dt} \end{eqnarray}$$(3) Figure 4. Open in new tabDownload slide Trend in renting out, agricultural participation, and agricultural land ownership. YearsSinceProgram|$_{dt,l}$| is an indicator that equals 1 if it has been l years since the start of the program in district d and year t; the omitted category is l = −1, or the year just before the program starts in any district. Due to the limits on the time periods covered by the survey data and the staggered timing of program start, the lags and leads relative to start date represented in the survey data can vary for the program start timing groups. For instance, for a district in the 2012 timing group, the survey data represents the following lags and leads with respect to program start: −7, −6, −5, −4, −2, −1, 0, 1, 2, and 3. Similarly, for the 2013 timing group, the lags range from −8 to 2 (−4 is missing). Thus, in specification (3), each YearsSinceProgram dummy coefficient would be driven by a different set of districts. To keep the sample of districts mostly stable, I show six lags and two leads. To account for the missing year, I follow McCrary (2007) and interpolate linearly between observation years correcting the standard errors for the induced serial correlation. With the two surveys combined, the data ranges from 2005 to 2015 with nine time fixed effects ranging from −6 to +2 for a balanced sample of districts.33 The graphs show that the program start is not driven by changes in land market activity, as the trend is flat in the pre-program period. In the post-program period, land rental increases, while agricultural participation declines. Land owned shows a flat pre- and post-program trend as reflected in the regression analysis earlier. 5.4. Placebo Program Rollout As an alternate test to rule out that pre-existing trends in the main outcomes drive the rollout of the program, I construct a placebo variable to measure “program intensity”, assuming the program rollout began two survey years prior to the actual program date in each district.34 Online Appendix Table A.22 provides the outcomes from specification (1) replacing intensity with the placebo treatment and shows no effect on the main outcomes. 5.5. Standard timing DD estimation I use two alternate identification strategies to measure the program effects, a standard timing DD and a stacked DD. First, I use a dummy variable, PostProgram, indicating the years after one of the subdistricts in a district has received the program in Online Appendix Table A.23. This treatment classification would avoid any concerns about endogenous pace of program delivery after the first opening in any district. The PostProgram indicator specifically captures the average effect of one subdistrict receiving the program (while the ProgramIntensity coefficient in the primary specification captures the effect of all subdistricts receiving the program); thus, the effects may be smaller. Online Appendix Table A.23 shows a significant increase in land rental by landowners and a significant drop in agricultural participation, while cultivating households have significantly more area rented in for cultivation and higher farm output. The magnitude of the effects is smaller, and the effect on cultivated area is positive but loses statistical significance. Goodman-Bacon (2018) cautions about an important feature of timing DD strategies with early and late timing groups. In particular, the comparisons of late timing groups to early timing groups rely on a comparison of just treated groups to already treated groups. If the early treated units are set on a differential trend by the treatment, they are no longer good “control” units. Goodman-Bacon (2018) proposes a decomposition to calculate the weight on each DD estimate. In my context, a calculation of the weights shows that the DD estimates based on the comparisons of early to late units have the majority of the weight (70%) while the DD effects comparing newly treated units to already treated ones have a much lower weight (30%). Moreover, the average effects from the two types of comparisons are qualitatively similar for all the main outcomes. Thus, the timing DD estimates are a meaningful comparison for the effects from the main regressions. 5.6. Stacked DD Estimation Another strategy uses a stacked DD to compare different timing groups to “control” units that are treated in a future period but are untreated when they act as controls, as in Deshpande and Li (2019). In particular, I construct a data set as follows: for each center opening, I label the district with the opening as treated and the districts that do not have any opening yet (but will receive the program one to two years after the treated district) as control. For each opening, I also include outcomes for the treated and control districts from survey rounds in the year before and just after opening to capture the change in trend due to the opening. The time period just after the opening is indicated by a Post indicator. I construct the treated and control districts for each of the openings and stack these datasets. A district is treated when the opening corresponds to a service center in one of its subdistricts, and a district could be treated for some openings and control for others. However, treated districts never switch to being controls because the set of control districts corresponds to districts that have not yet been treated. The specification I run is as follows: $$\begin{eqnarray} y_{{idt,o}}&=&\phi _{0}+\phi _{1}{{Treated}}_{d,o} \times {{Post}}_{t,o}+ \phi _{2}{{Post}}_{t,o}+\omega _{d,o}+ X^{\prime }_{{idt}} \Psi\nonumber\\ && +\;\mu _{d}+\eta _{t}+\epsilon _{{idt,o}}, \end{eqnarray}$$(4) where y|$_{idt}$| is an outcome for household i in district d, year t for opening o. Treatedd,o is 1 if d is the district where opening o occurred. Postt,o is 1 for the year after opening o occurs. Year fixed effects are included, as well as the same household controls from the main specification. Additionally an opening by district fixed effect is included, which is collinear with Treatedd,o. The coefficient on Treatedd,o × Postt,o captures the shift in outcome y just after the center opening in the treated district versus control districts. Online Appendix Table A.24 shows the outcomes from this empirical specification. In comparison to the main treatment effects that measure the effect of center openings in all subdistricts of a district, the coefficient in Online Appendix Table A.24 captures the effect of an opening in one subdistrict. These effects are naturally smaller in magnitude, but are significant and consistently in line with the main treatment effects. The effects on farm area and output are less precisely estimated but have the expected magnitude. In summary, the two additional empirical strategies provide reassuring complementary evidence to the established treatment effects above. 5.7. Multiple Hypotheses Testing Correction I implement Anderson’s (2008) methods to correct my standard errors for multiple hypotheses testing, controlling for the false discovery rate of Benjamini and Hochberg (1995) following Banerjee et al. (2015) and Ksoll et al. (2016) (Online Appendix Table A.25). The program effects on the primary outcomes, including land rental, agricultural participation by landowners, the total farm size, rental land size, and total output of cultivating households, survive this adjustment of p-values. Together, the robustness checks validate the identification strategy and provide confirmation of the program effects documented in the primary empirical analysis. 6. Concluding Remarks I focus on a “light touch” property rights formalization program in a context with private ownership, without explicit titling or direct targeting of market transactions. Roth and McCarthy (2013) note a continuum of land rights formalization that extends from strengthening tenurial rights in law or formal titling and registration to better communicating those rights to land holders or strengthening informal land leasing arrangements and contracts. The Punjab land record computerization program resulted in a formalization of property rights through better clarity of and access to rights, and through automation of market transactions bypassing bureaucratic hurdles and corrupt officers. The formalization of property rights can have potentially large positive effects while obviating the financial and feasibility hurdles of titling programs. The paper offers evidence that the program managed to significantly affect land markets, affecting allocation of land within agriculture and selection of cultivators into agriculture. Landowners who faced market constraints rent out land and exit agriculture after the program. On the flip side, households that stay in cultivation, rent in more land, effectively increasing average farm size, which has implications for modernization and aggregate agricultural productivity. Consistent with the increased rental activity and improved land allocation, aggregate yield improves in districts with the program, although the yield effects are not observed in farm-level data. I provide additional evidence that these changes in market activity are driven by improved security of tenure and verifiability of land rights. The results thus illustrate that land and labor market constraints limit rural mobility in the South Asian context, shedding light on the rural–urban divide and the prospect of structural transformation. The paper further reinforces our understanding of development economics by exhibiting how ICT use is manifested in public service processes and can ease market frictions in lower income countries. Effective use of ICT has been demonstrated for agricultural initiatives (Aker, Ghosh, and Burrell 2016), delivering education and improving learning (Muralidharan, Singh, and Ganimian 2019; Beg et al. 2020), increasing service delivery staff accountability (Duflo, Hanna, and Ryan 2012), and reducing leakages in government welfare program payments (Banerjee et al. 2014; Muralidharan, Niehaus, and Sukhtankar 2016). The land record computerization program similarly improves access to property rights records through digitized and automated land record maintenance. Notes The editor in charge of this paper was Imran Rasul. Acknowledgments I thank Imran Rasul (managing editor) and three anonymous referees for extensive comments. I am also grateful for useful feedback from Chris Udry, Marc Bellemare, Adrienne Lucas, Emily Oster, Laura Schechter, Shing-Yi Wang, James Fenske, Jeremy Tobacman, Ajay Shenoy, James Berry, Richard Hornbeck, and seminar participants at various seminars and conferences. I am grateful to Ghazala Mansuri and the Punjab Land Records Management Information System for useful data. Attique Ur Rehman, Hana Zahir, and Muhammad Zubair Rafique provided excellent research assistance. All errors are my own. Footnotes 1. Gollin, Parente, and Rogerson (2002). 2. A patwari was a historically appointed officer during the British colonial government, and has persisted as an office in the present land management system. 3. There is a 34-fold difference in average farm size (land per farm) between rich and poor countries. 4. For instance, the extent of inefficiency is larger on farms without marketed land in Malawi (Restuccia and Santaeulalia-Llopis 2017). 5. Chen (2017) offers additional theoretical support by demonstrating that untitled land cannot be traded across farmers, creating land misallocation and distorting individuals’ occupational choice between farming and working outside agriculture. 6. See Field (2007), Do and Iyer (2008), Galiani and Schargrodsky (2010), Deininger, Ali, and Alemu (2011), Ali, Deininger, and Goldstein (2014), Feder (1988), Besley (1995), Goldstein and Udry (2008), and Hornbeck (2010). 7. Deininger, Ali, and Alemu (2010) and Lunduka, Holden, and Øygard (2010) provide evidence suggesting tenurial insecurity prevents the efficient functioning of the land rental market in Ethiopia and Malawi. Macours, de Janvry, and Sadoulet (2010) find that tenurial insecurity constrains the matching of landlords and tenants in Nicaragua, affecting contractual outcomes. 8. de Janvry et al. (2015) find that land certificates in Mexico increase the likelihood of households to have a migrant member. 9. For India, Pakistan, and Bangladesh, it accounts for 17%, 25%, and 16%, respectively. Consistent with the high participation in agriculture, the average proportion of rural population in South Asia is 67% of the total, a decrease since 1960 but a much slower decline compared to Latin America. 10. Background about land record documents is based on United Nations Human Settlements Programme (2012). 11. The remainder report paying an illegal fee or do not respond as this payment is illegal by design. 12. Even though the patwar’s role is not abolished, approximately 150 service centers took on the tasks provided previously by approximately 8,000 patwaris. 13. Excludes tribal tehsils. 14. In addition to the HIES, the Pakistan Social and Living Standards Measurement (PSLM) surveys interview 80,000 households (nationally) and collect information on demographics, employment, access to public services, and key social indicators. The HIES has a larger questionnaire and smaller sample, while the PSLM has a larger sample but does not contain key farm-related data. For this reason, I use the HIES in the main regressions, but show additional analysis in the Online Appendix using data from the PSLM for outcomes measured in both surveys. 15. There is significant overlap between the landowning and cultivating sample, as approximately 80% of the landowning sample also cultivates land. 16. Assets include agricultural land, non-agricultural land, commercial land, residential land, cash savings, precious metals, and financial assets. 17. Using these input shares assumes farming all other crops is similar to rice farming, and the TFP residual is not over- or under-estimated due to differences in input usage for farming other crops. To test for robustness, I also use the input elasticities for all crops calculated in Chari et al. (2020) and find that the TFPs from the two methods have a correlation of 0.97. In the later analysis, I rank farmers in a district by their productivity and examine how the program affects allocation of land across this ranking. The results are qualitatively identical when I use the alternate input shares for productivity calculation. 18. Correlates of farmer TFP are discussed in Online Appendix Section D. 19. I use the 2011 and 2013 survey rounds for these tests, since the variation in |${{Program\_Intensity}}$| is primarily during those rounds. 20. For instance, a 1 standard deviation higher level of land rental is associated with a 0.02 percentage point higher program intensity in the following period. Thus, the difference in land rental rate across districts has almost no relation with the program’s rollout. 21. Districts belong to four possible timing groups depending on when the program starts in any district (2011, 2012, 2013, or 2014), and the timing DD estimate is a weighted average of two types of DD estimators (Goodman-Bacon 2018). The first type compares the change in outcomes for early districts (treated) to late districts (control) before and after the start of the program. The second type compares the change in outcomes for late districts to early districts before and after the treatment starts for the late districts (while the early districts have already been treated for some time). Online Appendix Figure A.1 illustrates two possible DD estimators when there are two timing groups. However, Goodman-Bacon (2018) points out concerns with the standard timing DD if the treatment effects vary over time. In this context, the treatment effect may increase as the program expands to the other subdistricts after it is initiated at the district level. Thus, the preferred specification uses the intensity variable that accounts for subsequent openings after program start in a district. 22. Based on an independent report of the land records service centers’ usage. 23. This usage rate measures any visit since the service center has been open (a four to eight year period) while the administrative visit data above measure visits within one calendar year. 24. There is also no significant effect on the size of owned holdings. 25. A natural outcome to test would be the rate of migration. The data do not allow us to test this explicitly, but the demonstrated effects suggest migration may have increased for landowning households. We can test if households are more likely to receive remittances or participate in the credit market. These outcomes are shown in Online Appendix Table A.6, which demonstrates no significant effects on the likelihood of a loan or the likelihood of receiving remittance income. The table also shows that total income for landowning households improves by 22% due to improved land and labor markets with the rollout of the program. 26. Online Appendix Table A.9 shows that the extensive margin of renting in among cultivating households does not change due to the program. As the composition of the sample of cultivating households may change due to the program (as landowners quit cultivation), I check for robustness by controlling for an indicator for land ownership by the household (Online Appendix Table A.10). 27. Allocative efficiency arises from moving land to high MPL farmers. In the pre-reform data, TFP and MPL are positively correlated, as may be expected in markets with frictions in land leasing or transfers. 28. The lowest TFP farmers may choose to rent out all their land and exit agriculture entirely. Since the cultivation data are used to calculate farmer TFP, households that do not participate in cultivation are excluded from the above regressions and I cannot directly test if lowest TFP households exit cultivation. 29. It can be noted that MPL is directly proportional to farm yield—MPL dispersion would simply capture the dispersion in yield. 30. Value-added or profits are calculated as the difference in the value of output per acre and the total expenses per acre. 31. Cereals: rice, wheat, and maize. Cash crops: cotton and sugarcane. 32. The proxy assumes that farmers who do not have rights to sell their property report no expected payment or an expected payment of zero. 33. The districts that start the program in 2014 are not included in the estimation of the coefficient for YearsSinceProgram = +2. 34. This robustness exercise yields similar results if I assume the placebo program started in the prior survey year. References Adamopoulos Tasso , Restuccia Diego ( 2014 ). “ The Size Distribution of Farms and International Productivity Differences .” American Economic Review , 104 ( 6 ), 1667 – 97 . Google Scholar Crossref Search ADS WorldCat Adamopoulos Tasso , Brandt Loren, Leight Jessica, Restuccia Diego ( 2017 ). “ Misallocation, Selection and Productivity: A Quantitative Analysis with Panel Data from China .” NBER Working Paper No. 23039 , NBER . Aker Jenny C. , Ghosh Ishita, Burrell Jenna ( 2016 ). “ The Promise (and Pitfalls) of ICT for Agriculture Initiatives .” Agricultural Economics , 47 , 35 – 48 . Google Scholar Crossref Search ADS WorldCat Ali Daniel Ayalew , Deininger Klaus, Goldstein Markus ( 2014 ). “ Environmental and Gender Impacts of Land Tenure Regularization in Africa: Pilot Evidence from Rwanda .” Journal of Development Economics , 110 , 262 – 275 . Google Scholar Crossref Search ADS WorldCat Anderson Michael L. ( 2008 ). “ Multiple Inference and Gender Differences in the Effects of Early Intervention: A Reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects .” Journal of the American Statistical Association , 103 , 1481 – 1495 . Google Scholar Crossref Search ADS WorldCat Banerjee Abhijit , Duflo Esther, Imbert Clement, Mathew Santhosh, Pande Rohini ( 2014 ). “ Can E-Governance Reduce Capture of Public Programs? Experimental Evidence from a Financial Reform of India’s Employment Guarantee .” Technical Report , MIT , December 2014 . Banerjee Abhijit , Duflo Esther, Goldberg Nathanael, Karlan Dean, Osei Robert, Parienté William, Shapiro Jeremy, Thuysbaert Bram, Udry Christopher ( 2015 ). “ A Multifaceted Program Causes Lasting Progress for the Very Poor: Evidence from Six Countries .” Science , 348 , 1260799 . Google Scholar Crossref Search ADS PubMed WorldCat Banerjee Aneesh , Jain Sachin ( 2003 ). “ E-Governance in India: Models that can be Applied in Other Developing Countries .” In Proceedings of the International Conference on Electronic Government . Springer , pp. 488 – 91 . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Beg Sabrin A. , Lucas Adrienne M., Halim Waqas, Saif Umar ( 2020 ). “ Engaging Teachers with Technology Increased Achievement, Bypassing Teachers Did Not .” NBER Working Paper No. 25704 , NBER . Bellemare Marc F. ( 2012 ). “ Insecure Land Rights and Share Tenancy: Evidence from Madagascar .” Land Economics , 88 , 155 – 180 . Google Scholar Crossref Search ADS WorldCat Bellemare Marc F. ( 2013 ). “ The Productivity Impacts of Formal and Informal Land Rights: Evidence from Madagascar .” Land Economics , 89 , 272 – 290 . Google Scholar Crossref Search ADS WorldCat Benjamini Yoav , Hochberg Yosef ( 1995 ). “ Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing .” Journal of the Royal Statistical Society: Series B (Methodological) , 57 , 289 – 300 . Google Scholar OpenURL Placeholder Text WorldCat Besley Timothy ( 1995 ). “ Property Rights and Investment Incentives: Theory and Evidence from Ghana .” Journal of Political Economy , 103 , 903 – 937 . Google Scholar Crossref Search ADS WorldCat Besley Timothy , Ghatak Maitreesh ( 2010 ). “ Property Rights and Economic Development .” In Handbook of Development Economics , vol. 5 . Elsevier , pp. 4525 – 4595 . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Binswanger-Mkhize Hans P. ( 2013 ). “ The Stunted Structural Transformation of the Indian Economy .” Economic and Political Weekly , 48 , 5 – 13 . Google Scholar OpenURL Placeholder Text WorldCat Bloom Nicholas , Garicano Luis, Sadun Raffaella, Van Reenen John ( 2014 ). “ The Distinct Effects of Information Technology and Communication Technology on Firm Organization .” Management Science , 60 , 2859 – 2885 . Google Scholar Crossref Search ADS WorldCat Bresnahan Timothy F. , Brynjolfsson Erik, Hitt Lorin M. ( 2002 ). “ Information Technology, Workplace Organization, and the Demand for Skilled Labor: Firm-Level Evidence .” The Quarterly Journal of Economics , 117 , 339 – 376 . Google Scholar Crossref Search ADS WorldCat Carter Michael R. , Olinto Pedro ( 1996 ). Decomposition and Identification of the Impact of Land Titling on Agricultural Productivity in Paraguay . Department of Agricultural Economics, University of Wisconsin, Madison, WI . Chari Amalavoyal , Liu Elaine M., Wang Shing-Yi, Wang Yongxiang ( 2020 ). “ Property Rights, Land Misallocation, and Agricultural Efficiency in China .” The Review of Economic Studies . NBER Working Paper No. 24099 , NBER . Chen Chaoran ( 2017 ). “ Untitled Land, Occupational Choice, and Agricultural Productivity .” American Economic Journal: Macroeconomics , 9 , 91 – 121 . Google Scholar OpenURL Placeholder Text WorldCat Chen Chaoran , Restuccia Diego, Santaeulàlia-Llopis Raül ( 2021 ). “ The Effects of Land Markets on Resource Allocation and Agricultural Productivity .” NBER Working Paper No. 24034 , NBER . Colin Cameron A. , Miller Douglas L. ( 2015 ). “ A Practitioner’s Guide to Cluster-Robust Inference .” Journal of Human Resources , 50 , 317 – 372 . Google Scholar Crossref Search ADS WorldCat de Janvry Alain , Emerick Kyle, Gonzalez-Navarro Marco, Sadoulet Elisabeth ( 2015 ). “ Delinking Land Rights from Land Use: Certification and Migration in Mexico .” American Economic Review , 105 ( 10 ), 3125 – 49 . Google Scholar Crossref Search ADS WorldCat de Janvry Alain , Fafchamps Marcel, Sadoulet Elisabeth ( 1991 ). “ Peasant Household Behaviour with Missing Markets: Some Paradoxes Explained .” The Economic Journal , 101 , 1400 – 1417 . Google Scholar Crossref Search ADS WorldCat Deininger Klaus , Goyal Aparajita ( 2012 ). “ Going Digital: Credit Effects of Land Registry Computerization in India .” Journal of Development Economics , 99 , 236 – 243 . Google Scholar Crossref Search ADS WorldCat Deininger Klaus , Ali Daniel Ayalew, Alemu Tekie ( 2010 ). “ Land Rental Markets: Transaction Costs and Tenure Insecurity in Rural Ethiopia .” In The Emergence of Land Markets in Africa , pp. 70 – 86 . Routledge . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Deininger Klaus , Ayalew Ali Daniel, Alemu Tekie ( 2011 ). “ Impacts of Land Certification on Tenure Security, Investment, and Land Market Participation: Evidence from Ethiopia .” Land Economics , 87 , 312 – 334 . Google Scholar Crossref Search ADS WorldCat Deshpande Manasi , Li Yue ( 2019 ). “ Who is Screened Out? Application Costs and the Targeting of Disability Programs .” American Economic Journal: Economic Policy , 11 , 213 – 48 . Google Scholar OpenURL Placeholder Text WorldCat Didan K. ( 2015 ). MYD13Q1 MODIS/Aqua Vegetation Indices 16-Day L3 Global 250m SIN Grid V006 . LAADS DAAC , Greenbelt, MA . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Do Quy-Toan , Iyer Lakshmi ( 2008 ). “ Land Titling and Rural Transition in Vietnam .” Economic Development and Cultural Change , 56 , 531 – 579 . Google Scholar Crossref Search ADS WorldCat Duflo Esther , Hanna Rema, Ryan Stephen P. ( 2012 ). “ Incentives Work: Getting Teachers to Come to School .” American Economic Review , 102 ( 4 ), 1241 – 78 . Google Scholar Crossref Search ADS WorldCat FAO/IIASA/ISRIC/ISSCAS/JRC ( 2012 ). “ Harmonized World Soil Database (version 1.2) .” FAO, Rome and IIASA , Laxenburg . Google Scholar Feder Gershon ( 1988 ). Land Policies and Farm Productivity in Thailand . Johns Hopkins University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Field Erica ( 2007 ). “ Entitled to Work: Urban Property Rights and Labor Supply in Peru .” The Quarterly Journal of Economics , 122 , 1561 – 1602 . Google Scholar Crossref Search ADS WorldCat Field Erica , Torero Maximo ( 2006 ). “ Do Property Titles Increase Credit Access among the Urban Poor? Evidence from a Nationwide Titling Program .” Working Paper . Foster Andrew D. , Rosenzweig Mark R. ( 2011 ). “ Are Indian Farms Too Small? Mechanization, Agency Costs, and Farm Efficiency .” Brown University and Yale University . Google Scholar Foster Andrew D. , Rosenzweig Mark R. ( 2017 ). “ Are there Too Many Farms in the World? Labor-Market Transaction Costs, Machine Capacities and Optimal farm Size .” NBER Working Paper No. 23909 , NBER . Galiani Sebastian , Schargrodsky Ernesto ( 2010 ). “ Property Rights for the Poor: Effects of Land Titling .” Journal of Public Economics , 94 , 700 – 729 . Google Scholar Crossref Search ADS WorldCat Ghosh Atanu , Banerjee Gargi ( 2006 ). “ A Study of E-Governance in Rural India .” In Delivering E-Government , Global Institute of Flexible Systems Management. New Delhi , p. 222 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Goldstein Markus , Udry Christopher ( 2008 ). “ The Profits of Power: Land Rights and Agricultural Investment in Ghana .” Journal of Political Economy , 116 , 981 – 1022 . Google Scholar Crossref Search ADS WorldCat Gollin Douglas , Parente Stephen, Rogerson Richard ( 2002 ). “ The Role of Agriculture in Development .” American Economic Review , 92 ( 2 ), 160 – 164 . Google Scholar Crossref Search ADS WorldCat Gonzalez Mary Lisbeth ( 2016 ). “ Land Records Go Digital in Punjab, Pakistan .” https://blogs.worldbank.org/endpovertyinsouthasia/land-records-go-digital-punjab-pakistan, Accessed July 28, 2021 . Goodman-Bacon Andrew ( 2018 ). “ Difference-in-Differences with Variation in Treatment Timing .” NBER Working Paper No. 25018 , NBER . Gottlieb Charles , Grobovšek Jan ( 2019 ). “ Communal Land and Agricultural Productivity .” Journal of Development Economics , 138 , 135 – 152 . Google Scholar Crossref Search ADS WorldCat Hornbeck Richard ( 2010 ). “ Barbed Wire: Property Rights and Agricultural Development .” Quarterly Journal of Economics , 125 , 767 – 810 . Google Scholar Crossref Search ADS WorldCat Hsieh Chang-Tai , Klenow Peter J. ( 2009 ). “ Misallocation and Manufacturing TFP in China and India .” The Quarterly Journal of Economics , 124 , 1403 – 1448 . Google Scholar Crossref Search ADS WorldCat Jacoby Hanan G. , Mansuri Ghazala ( 2008 ). “ Land Tenancy and Non-Contractible Investment in Rural Pakistan .” The Review of Economic Studies , 75 , 763 – 788 . Google Scholar Crossref Search ADS WorldCat Ksoll Christopher , Bie Lilleør Helene, Helth Lønborg Jonas, Dahl Rasmussen Ole ( 2016 ). “ Impact of Village Savings and Loan Associations: Evidence from a Cluster Randomized Trial .” Journal of Development Economics , 120 , 70 – 85 . Google Scholar Crossref Search ADS WorldCat López Ramón , Romano Claudia ( 2000 ). “ Rural Poverty in Honduras: Asset Distribution and Liquidity Constraints .” In Rural Poverty in Latin America , pp. 227 – 243 . Springer . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Lunduka Rodney , Holden Stein T., Øygard Ragnar ( 2010 ). “ Land Rental Market Participation and Tenure Security in Malawi .” In The Emergence of Land Markets in Africa , pp. 125 – 143 . Routledge . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Macours Karen , de Janvry Alain, Sadoulet Elisabeth ( 2010 ). “ Insecurity of Property Rights and Social Matching in the Tenancy Market .” European Economic Review , 54 , 880 – 899 . Google Scholar Crossref Search ADS WorldCat McCrary Justin ( 2007 ). “ The Effect of Court-Ordered Hiring Quotas on the Composition and Quality of Police .” American Economic Review , 97 ( 1 ), 318 – 353 . Google Scholar Crossref Search ADS WorldCat Muralidharan Karthik , Niehaus Paul, Sukhtankar Sandip ( 2016 ). “ Building State Capacity: Evidence from Biometric Smartcards in India .” American Economic Review , 106 ( 10 ), 2895 – 2929 . Google Scholar Crossref Search ADS WorldCat Muralidharan Karthik , Singh Abhijeet, Ganimian Alejandro J. ( 2019 ). “ Disrupting Education? Experimental Evidence on Technology-Aided Instruction in India .” American Economic Review , 109 ( 4 ), 1426 – 60 . Google Scholar Crossref Search ADS WorldCat Newman Carol , Tarp Finn, Van Den Broeck Katleen ( 2015 ). “ Property Rights and Productivity: The Case of Joint Land Titling in Vietnam .” Land Economics , 91 , 91 – 105 . Google Scholar Crossref Search ADS WorldCat Restuccia Diego , Santaeulalia-Llopis Raul ( 2017 ). “ Land Misallocation and Productivity .” NBER Working Paper No. 23128 , NBER . Rosenzweig Mark R. ( 2001 ). “ Savings Behaviour in Low-Income Countries .” Oxford Review of Economic Policy , 17 , 40 – 54 . Google Scholar Crossref Search ADS WorldCat Roth Michael , McCarthy Nancy ( 2013 ). Land and Tenure, Property Rights and Economic Growth in Rural Areas . USAID Issue Brief . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Shenoy Ajay ( 2017 ). “ Market Failures and Misallocation .” Journal of Development Economics , 128 , 65 – 80 . Google Scholar Crossref Search ADS WorldCat Sural Ajay ( 2013 ). “ Punjab May Abolish Patwari, Kanungo Systems .” Times of India . Google Scholar United Nations Human Settlements Programme ( 2012 ). “ Land Tenure and Property Rights in Pakistan .” https://www.land-links.org/issue-brief/land-tenure-and-property-rights-in-pakistan/, retrieved 20 July 2021 . Wang Shing-Yi ( 2012 ). “ Credit Constraints, Job Mobility, and Entrepreneurship: Evidence from a Property Reform in China .” Review of Economics and Statistics , 94 , 532 – 551 . Google Scholar Crossref Search ADS WorldCat World Bank ( 2013 ). World Development Indicators 2013 . World Bank Publications . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC © The Author(s) 2021. Published by Oxford University Press on behalf of European Economic Association. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model) © The Author(s) 2021. Published by Oxford University Press on behalf of European Economic Association.
TI - Digitization and Development: Property Rights Security, and Land and Labor Markets
JF - Journal of the European Economic Association
DO - 10.1093/jeea/jvab034
DA - 2022-02-16
UR - https://www.deepdyve.com/lp/oxford-university-press/digitization-and-development-property-rights-security-and-land-and-gFcl16f8U7
SP - 395
EP - 429
VL - 20
IS - 1
DP - DeepDyve
ER -