Policy Research Working Paper 10140 Measuring Land Tenure at the Individual Level Lessons from Methodological Research in Armenia Sydney Gourlay Giuseppe Maggio Anahit Safyan Alberto Zezza Development Economics Development Data Group August 2022 Policy Research Working Paper 10140 Abstract Evidence indicates that land rights are strongly associated research has shown that the use of proxy respondents in with several indicators of well-being and development the collection of data on assets, including land, results in outcomes, including access to credit, resilience to shocks, a biased understanding of men’s and women’s holdings productivity, and bargaining power. Accurately captur- vis-à-vis self-reporting. This paper uses data from a meth- ing gender differences in land rights is thus critical for odological experiment in Armenia to assess the implications development policy, prompting the need to shift from of survey design—namely, respondent strategy and the level household-level land rights data collection to collecting of disaggregation of land data—on the measurement of more and better individual-level data on land rights. The individual land rights and SDG indicator monitoring. The importance of individual land rights has been recognized findings suggest that in the context of Armenia, the mea- in the Sustainable Development Goals (SDG) agenda, with surement of SDG 5.a.1 and 1.4.2 (a) is robust to respondent the inclusion of two key indicators on land rights—SDG approach and data disaggregation level, driven largely by indicators 1.4.2 and 5.a.1. Although clear guidance exists the high rates of documentation. Meanwhile, land rights for computing and monitoring these, the choice of data col- that are less objective, such as the right to bequeath and lection methods may influence the resulting indicators and perception of tenure security, are sensitive to these survey the understanding of the underlying land rights. Specifically, design choices. This paper is a product of the Development Data Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at sgourlay@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Measuring Land Tenure at the Individual Level: Lessons from Methodological Research in Armenia Sydney Gourlay1† Giuseppe Maggio†* Anahit Safyan‡ Alberto Zezza† JEL Classification: C8, J16, Q15 Keywords: land tenure; survey methodology; gender; Sustainable Development Goals; household surveys 1 Corresponding author: s.gourlay@worldbank.org. †Living Standards Measurement Study (LSMS), Data Production and Methods Unit, Development Data Group, World Bank, Rome, Italy. *Department of Law, Università di Palermo, Palermo, Italy. ‡International Statistical Cooperation Sphere, State Council on Statistics of the Republic of Armenia, Statistical Committee of the Republic of Armenia, Yerevan, Armenia. This is a publication of the 50x2030 Initiative to Close the Agricultural Data Gap, a multi-partner program that seeks to bridge the global agricultural data gap by transforming data systems in 50 countries in Africa, Asia, the Middle East, and Latin America by 2030. For more information on the Initiative, visit 50x2030.org. The authors thank Heather Moylan and Jaap Zevenbergen for their valuable comments on an earlier draft, Raka Banerjee for her editorial contribution, and the Statistical Committee of the Republic of Armenia and the ICARE Foundation for their hard work in survey implementation. Any remaining errors are the authors' sole responsibility. I. Introduction The rights to land ownership and land use are a critical component of development policy discussions, and are associated with increased access to credit, productivity, and resilience in the face of various shocks (Carter and Barrett, 2006). Recent work by Deininger et al. (2021) finds that land rights are associated with increased investment and cash crop adoption, and that the impact of these rights differs across men and women. Evidence also suggests that the ownership of assets like land at the individual level (as opposed to at the household level) is positively associated with intrahousehold bargaining power (Doss, 2013). Given the importance of land tenure to development outcomes, as well as evidence that the impacts of land tenure rights differ by gender, accurate measurement and monitoring of land rights is essential for informed policy. The Sustainable Development Goal (SDG) agenda recognizes the importance of land tenure by including several indicators on individual land rights. As the SDG framework strives to provide a dashboard-type look at development progress and outcomes, complex concepts such as land tenure are necessarily distilled into simplified indicators that can be widely and regularly reported. Taking the SDG indicators related to land tenure as defined in the indicator metadata, we provide evidence that the choice of measurement method may alter indicator estimations, particularly given the various nuances of land tenure that are not readily captured in the simplified indicator framework. SDG indicators 1.4.2 and 5.a.1, which fall under the goals of ending poverty and achieving gender equality, respectively, both focus on individual, sex-disaggregated land tenure rights. Indicator 1.4.2 assesses tenure security for the population at large, based on documented rights and perceptions of security, while indicator 5.a.1 monitors land tenure security for the agricultural population, through the concepts of documented rights and alienation rights. While all 193 United Nations member states have committed to pursuing and reporting on the SDGs, only 34 countries had either fully or partially reported on SDG 1.4.2, and only 33 countries on SDG 5.a.1, as of April 2022. 2 The majority of the countries reporting these indicators have done so using estimated data, such as using proxies for the codified definitions of the indicators, using a subsample of the relevant population, or reporting on a subset of the sub-indicators due to constraints around data availability. Given the heavy data burden of the SDG agenda for national statistical agencies, it is imperative that affordable, efficient, and scalable tools are readily available. The custodian agencies for indicator 1.4.2 – UN-Habitat and the World Bank – and indicator 5.a.1 – the Food and Agriculture Organization of the United Nations (FAO) – worked with stakeholders such as Landesa and the Evidence and Data for Gender Equality (EDGE) initiative to develop a questionnaire module for jointly collecting the data for both indicators (FAO, World Bank, and UN-Habitat, 2019). The questionnaire module was designed to facilitate widespread uptake by allowing for integration into various types of surveys, from detailed household surveys with extensive agricultural modules to specialized household surveys with limited scope for land data collection. To accommodate a wide variety of data collection efforts, multiple versions of this module were developed, differentiated over two characteristics: the respondent strategy and the level of data collection. On the respondent strategy, modules were designed for both a self- respondent approach (where one or more randomly selected respondents answer only for themselves) and a proxy respondent approach (where a given individual answers on behalf of themselves as well as other household members). On the level of data collection, modules were designed to either collect data for each parcel separately (parcel-level data), which allows for disaggregation of the indicators by type of land tenure, or for all land in aggregate (aggregate-level data). For surveys that do not already collect parcel-level data, which is often the case 2 Global SDG Indicators Database, accessed April 26, 2022: https://unstats.un.org/sdgs/indicators/database/ 2 for surveys other than detailed agricultural surveys, the approach of aggregated land is likely a more practical and affordable option. Similarly, survey practitioners may lean towards the proxy respondent approach, given the apparent ease of implementation relative to the self-respondent approach. However, the ease of implementation of a given module must be weighed against the data quality implications associated with that choice. Moore (1988), in a literature review on survey response quality of self and proxy respondent approaches, highlighted the lack of studies isolating the impact of proxy response, calling on researchers to further explore this theory through studies carefully designed for this purpose. Since then, research on proxy respondent bias through methodological validation studies has been conducted across multiple sectors. Todorov and Kirchner (2000) found that the use of proxy respondents in representative surveys on disability resulted in systematically biased estimates of national disability rates, with proxy reports estimating 8.8% of people to be experiencing functional limitations, as compared to 16.9% indicated by self-reports (although the direction and magnitude of the bias differed across age group and disability type). In a study on the measurement of income using data from Malawi, Fisher, Reimer, and Carr (2010) found that in 66% of households, the husband underestimated his wife’s income by an average of 47%. This resulted in total household income being underestimated by 26% on average across the full sample when the husband reported income for himself and his wife, as compared to total household income measured through husband and wife self-reports. Kilic et al. (2020) explored the implications of proxy versus self-responding on measuring employment in Malawi, finding that allowing proxy respondents led to underestimates of employment in multiple sectors and with gender- differentiated magnitudes relative to private self-respondent interviews. The effect of proxy respondents on the measurement of assets in low-income contexts has recently been explored by the EDGE initiative and the Living Standards Measurement Study-Plus (LSMS+) project. 3 The United Nations Guidelines for Producing Statistics on Asset Ownership from a Gender Perspective (henceforth referred to as ‘UN Guidelines’; UNSD, 2019), based on empirical evidence amassed through the EDGE project, explicitly state: “individual-level data on asset ownership should be reported by self rather than proxy, owing to large discrepancies between proxy and self-responses…”. Evidence from the Uganda Methodological Survey Experiment on Measuring Asset Ownership from a Gender Perspective (MEXA), which fed into the development of the UN EDGE guidelines, illustrates the asymmetric impacts of respondent approach by gender. MEXA found that the use of self-reports increased women’s reported ownership of agricultural land by 5 percentage points relative to proxy reports, while the increase for men’s reported ownership was 10 percentage points (Kilic and Moylan, 2016; UNSD, 2019). In line with this and as part of the LSMS+ project, Kilic, Moylan, and Koolwal (2020) found that the use of proxy respondents relative to self-respondent implementation resulted in higher rates of exclusive ownership of agricultural land for men, and lower rates for women. Thus, these gender-differentiated biases based on respondent strategy may not only lead to inaccurate estimates of land tenure for the population at large, but they also have the potential to distort estimates of the gender gap in rights. On the relationship between data quality and level of data collection, there is a dearth of research on the implications of collecting land tenure data at the parcel level vis-à-vis the aggregate level. However, related research on the effects of aggregation raises concerns for data quality implications and motivates the need for methodological validation. Research by Beegle et al. (2012), for example, explored the impacts of the level of detail in a consumption questionnaire on the resulting consumption estimates. The a priori assumption is that greater levels of detail, specifically a longer list of specific consumption items, trigger respondents’ memories and prevent 3For reports on the LSMS+ surveys in Tanzania, Malawi, Ethiopia, and Cambodia, see Hasanbasri et al. (2021a) and Hasanbasri et al. (2021b). 3 the omission of certain categories. Beegle et al. (2012) found that using a collapsed, or aggregated, list of consumption items results in significantly lower estimates of total consumption relative to the gold standard for measuring consumption (a personal diary with frequent supervision). On the contrary, research on the measurement of microenterprise profits by de Mel, McKenzie, and Woodruff (2009) concluded that asking respondents about profits in aggregate, as opposed to asking several detailed questions on revenues and expenses, resulted in more precise estimates of profits. The degree and direction of bias introduced by the aggregation of land to the individual level in the measurement of tenure rights remains an open empirical question. Using data from the Armenia Land Tenure and Area (ALTA) study and in the context of land tenure rights, this paper addresses questions on: (i) the data quality implications associated with respondent strategy; (ii) the data quality implications associated with level of data collection; (iii) the interaction of biases stemming from the use of proxy respondents and aggregated level data collection, as relevant; and ultimately, (iv) the implications of these design decisions on the computation and monitoring of SDGs 1.4.2 and 5.a.1. The ALTA study was implemented with the technical guidance of the lead author by the International Centre for Agribusiness Research and Education (ICARE) in partnership with the Statistical Committee of the Republic of Armenia (ArmStat) and with the financial support of the 50x2030 Initiative. ALTA included 1,200 scientifically sampled households which were randomly assigned to one of four treatment arms, each of which received a specific version of the land tenure module differing by respondent approach and level of data collection. This experimental design provides a unique opportunity to identify the biases resulting from these particular survey design choices and the implications for SDG indicator computation. Findings from this study will inform future survey design around these particular SDG indicators and land tenure data in general, as well as alert policy makers to the implications of SDG reporting using various survey design approaches. To our knowledge, this is the first paper to address questions around the feasibility and validity of land tenure data collected in aggregate vis-à-vis the parcel level and will contribute to the growing literature on respondent approaches, particularly as related to individual land tenure. The paper is organized as follows: section II describes the land tenure system in Armenia, providing context for the results that follow; Section III describes the ALTA study design and data; Section IV discusses SDG indicator computation and methods; Section V presents the results; and Section VI concludes. II. Armenian Land Tenure System Following the dissolution of the Soviet Union in 1991, the Republic of Armenia became independent, implementing numerous reforms in the process of transitioning to a liberal economy. Agricultural land was considered to be the starting point for privatization, with subsequent reforms following suit in other sectors of the economy. The legal agrarian reforms were enacted through a number of laws, including "On Property in the Republic of Armenia" adopted by the Supreme Council of the Republic of Armenia (October 1990), "On Rural and Rural Collective Farms" (adopted January 1991), "On Enterprises and Entrepreneurial Activity in the Republic of Armenia" (adopted March 1992), “RA Land Code” (adopted January 1991), and several other laws and by-law acts. This legislation resulted in a new agrarian land tenure system beginning in March 1991 which converted the approximately 900 previously existing collective farms, state farms, and inter-economic enterprises into a large number of smaller family farms. The conversion of state land into private land was conducted in such a way that each village family was given a number of land units, dependent on the number of household members (one unit of land for families with three or fewer members, two units of land for families with four to six members, and 4 three units of land for families with seven or more members). The land units were selected across different types of land, and therefore families were generally allocated several fragmented parcels of land of varying types (Giovarelli & Bledsoe, 2001). Over 1.2 million separate agricultural parcels were registered with the State Committee of the Real Property Cadastre following the 1991 privatization (UNECE, 2005). While some state and collective farms persisted, with approximately 30% of arable land retained by the government under this system, a large share of that land was also leased to family farmers (Lerman & Mirzakhanian, 2001). Findings from the 2014 Agricultural Census suggest there are over 317,000 family farms 4 of 1.48 hectares on average (Statistical Committee of the Republic of Armenia, 2016), which produce approximately 95% of the country’s gross agricultural output (Statistical Committee of the Republic of Armenia, 2021). As part of the land tenure reforms, state-provided documents confirming the ownership of the given land were provided to the new landowners. Based on these documents, the Cadastre Committee provided certificates which granted the rights of the named owners to sell, rent, and bequeath the land. However, these documents were not provided to all newly established farms; in some cases, their ownership was registered in community registers or with village municipalities, which had no legal force. Based on the information recorded in the local registers, the communities or village municipalities, if necessary, provided documents to the farms that proved the ownership of the land by the farmer. These types of documents, however, did not allow the farmer to legally sell their land nor to rent it out. Given the land reforms described above, Armenian households generally have high rates of documented land rights. III. Data and Experiment Design The sample of the ALTA study consists of 1,200 households, scientifically selected from 100 urban and rural enumeration areas (EAs) covering both agricultural and non-agricultural households. All EAs fall within three Figure 1. ALTA Study Area marzes, namely Ararat (34 EAs), Kotayk (33 EAs), and Vayots Dzor (33 EAs), which were selected to capture variation in geographic terrain and land use, while also considering marz capacity for effectively implementing the methodological study (illustrated in Figure 1). EAs were not stratified within the selected marzes. The sample frame 4 This estimate of over 317,000 farms includes those farms that had their own land (including leased land) at the time of the 2014 Agricultural Census, as well as farms that leased land but did not have their own land. It excludes farms that were absent from the community at the time of the survey. 5 of enumeration areas was based on the Population Census 2011, and a fresh household listing was conducted in each of the selected EAs to attain a current household sampling frame from which to randomly select 12 households in each EA. Randomization of the treatment arm assignment was conducted within each EA, to account for unobserved differences at the village level that could potentially affect responses to the land tenure questions as well as to avoid confounding enumerator effects. From each EA, three households were assigned to each of the treatment arms identified in Table 1. Treatment arm assignment was completed at the headquarters level and preloaded in the computer assisted personal interviewing (CAPI) program developed in Survey Solutions, 5 and therefore could not be altered by enumerators. In Arms 1 and 2, up to three adult household members were randomly selected by the Survey Solutions software for individual interviews, again limiting the scope of bias in the ALTA sample. Table 1. ALTA Treatment Arms Respondent Level of Land Approach Data Collection Arm 1 Self-Respondent Parcel Arm 2 Self-Respondent Aggregate Arm 3 Proxy Parcel Arm 4 Proxy Aggregate The versions of the land tenure module, and therefore the treatment arms, differ on the level of respondent approach, either self- or proxy respondent, and in the level of land reporting, either parcel or aggregate level. Individuals selected for the self-respondent questionnaire report the land tenure status on the land they directly own or use, while proxy respondents answer the same set of questions for themselves as well as for other individuals in their household. In parcel-level treatment arms, questions on land tenure are asked for each reported parcel separately, while aggregate-level treatment arms ask about the land tenure status for the overall land owned or used by specific household members, with the only disaggregation being by agricultural and non-agricultural land use (as this is a necessary distinction for SDG 5.a.1). As Arm 1 entails both the highest level of disaggregation (parcel-level) and the presumed highest quality information (self-respondent) as advised by the UN Guidelines, we considered it to be the gold standard against which we compare the results from the other three arms. In addition to the module on land tenure, the ALTA questionnaire included a brief set of modules capturing individual socio-demographic characteristics, educational attainment, employment status, and agricultural activity. 6 An additional focus of the ALTA study was the validation of measurement approaches to land area estimation. To this end, both Arms 1 and 3 included a module on land area measurement that incorporated GPS and satellite imagery-based parcel-level area measurement. A series of cognitive interviews were conducted to ensure that the land tenure questions were translated as intended from English to Armenian, as well as to understand how the interpretation of questions varied across individuals, with a view to understanding male versus female 5 For more information on the publicly available Survey Solutions CAPI software, visit: https://mysurvey.solutions/en/ 6 The ALTA questionnaires are available on the LSMS website here: https://www.worldbank.org/en/programs/lsms/land- tenure. ALTA data will be made available in the World Bank’s Microdata Library by September 2022. 6 interpretation of the land-related questions. This cognitive interviewing exercise led to the refinement of the phrasing of certain questions to address both translation and respondent interpretation issues. Summary statistics are reported in Table 2. In the sample as a whole, 73% of households are deemed agricultural households, as defined by the SDG 5.a.1 methodology. Households, on average, are composed of four individuals, three of whom are adults (that is, 18 years or older). T-tests between Arm 1 and the other three arms indicate no significant differences in the characteristics presented in Table 2, with the exception of the age of the household head, which differs slightly between Arm 1 and the other arms. Table 2. Summary Statistics, by Arm Total Arm 1 Arm 2 Arm 3 Arm 4 HH size 4.0 4.0 3.8 4.1 3.9 # of adults/HH 3.0 3.1 2.9 3.1 3.0 Female headed HH (%) 0.31 0.31 0.30 0.28 0.34 Head age (years) 60.2 58.2 60.4** 60.7** 61.3*** # parcels owned/used per HH - 1.48 - 1.42 - Agricultural HH (%) 0.73 0.75 0.70 0.74 0.71 Urban (%) 0.36 0.36 0.36 0.36 0.36 Marz distribution (%) Ararat 0.34 0.34 0.34 0.34 0.34 Kotayk 0.33 0.33 0.33 0.33 0.33 Vayots Dzor 0.33 0.33 0.33 0.33 0.33 N (HHs) 1200 301 300 298 301 Note: Asterisks indicate the value is statistically different from the corresponding value for Arm 1 at the 1% (***), 5% (**), and (10% *) levels. IV. Indicator Concepts, Definitions, and Methods One of the main challenges when measuring land rights or tenure security is the definition in itself, as well as its operationalization. The concepts of land rights and land tenure security vary depending on the context of the analysis, leading to the emergence of a set of proxies employed throughout the literature to address questions of land rights (Doss et al., 2015). For example, studies analyzing the relationship between land rights and productivity or income in agriculture have focused more on the use right, which is associated with the individual who makes farming decisions and withdraws outputs from the land, usually denoted as the land manager (Bambio and Agha, 2018; Lawry et al., 2017; Asfaw and Maggio, 2018). On the other hand, literature investigating the role of land as collateral in credit markets or studying farmer reactions to the efforts of large-scale investors to consolidate land has generally focused on documented ownership, deriving from the land registration and the owner’s possession of official documents (Domeher and Abdulai, 2012; Zhang et al., 2020; Joel and Bergaly, 2020). Work by Hasanbasri et al. (2022) uses data from Ethiopia, Malawi, and Tanzania to illustrate the importance of measuring multiple dimensions of land tenure, as gender gaps can vary significantly by type of right. The following are commonly used concepts for the analysis of land rights, which may be used jointly or in isolation, depending on the objective of the analysis and the features of the land tenure context in which the analysis is being conducted: 7 - Reported ownership: “reported ownership refers to the persons who consider themselves to be owners of the asset in question, irrespective of whether they possess legal, or documented, ownership of the asset” (UN Guidelines). - Documented ownership: “documented ownership refers to the existence of any document recognized by the Government that an individual can use to claim ownership rights in law over an asset by virtue of the individual’s name being listed as an owner on the document” (UN Guidelines). - Legally recognized documentation: Based on the definition of documented ownership, but broadened to include legally recognized documents that provide evidence of use rights (such as rental contracts), legally recognized documentation refers to the existence of a legally recognized document in the name of a given individual that provides evidence of ownership or use rights. - Right to sell: “the right to sell an asset refers to the ability of an individual to transfer the asset in question permanently, in return for cash or in-kind benefits. This right may be held jointly with one or more individuals. The right to sell an asset is the right most commonly associated with ownership, but the concept is not applicable in areas where laws or social norms preclude the sale of assets, such as land” (UN Guidelines). - Right to bequeath: “the right to bequeath an asset refers to the ability of individuals to give the asset in question, by oral or written will, to other persons after their death. This right may be held jointly with one or more individuals. The right to bequeath… may be more universal than the right to sell, since in many contexts owners can bequeath assets to their children or other persons even if they are prohibited from selling them” (UN Guidelines). In some cases, such as in the questionnaire module for measuring SDG 1.4.2 and 5.a.1 and the associated guidance document (FAO, World Bank, and UN-Habitat, 2019), the right to bequeath refers to intergenerational transfer of land rights both in life and death. The metadata for SDG indicator 5.a.1 describes in detail the methodology and definitions underlying the indicator, as approved by the Inter-agency and Expert Group on SDG Indicators and adopted as the official methodology for monitoring the indicator. Specifically, the metadata describes the inclusion of legally recognized documentation, right to bequeath, and right to sell in the indicator construction. The metadata for indicator 1.4.2 explains that the concepts of legally recognized documentation and right to bequeath are also used in the computation of this indicator, in addition to one concept not noted above – the perception of tenure security. - Perception of tenure security: “Perception of tenure security refers to an individual’s perception of the likelihood of involuntary loss of land, such as disagreement of the ownership rights over land or ability to use it, regardless of the formal status and can be more optimistic or pessimistic” (SDG 1.4.2 metadata). Here, we focus on the manner in which the SDG indicators are constructed, in line with the formulations identified in the indicator metadata, followed by comparative analysis of the resulting indicator estimations across arms. Note that while the respondent approach and level of aggregation differs across the four versions of the land tenure module, the formulation of the questions has been kept consistent to ensure that the SDG indicators can be computed in the same manner. The indicators are defined as follows: 8 Indicator 1.4.2: “Proportion of total adult population with secure tenure rights to land, with (a) legally recognized documentation; and (b) who perceive their rights to land as secure, by sex and by type of tenure” Indicator 5.a.1: “(a) Proportion of total agricultural population with ownership or secure rights over agricultural land by sex; and (b) share of women among owners or rights bearers of agricultural land, by type of tenure” Under indicator 5.a.1, an individual in the agricultural population 7 is considered to have secure land rights if they have either the right to sell, the right to bequeath, or their name on a legally recognized document. Accordingly, the sub-indicators for 5.a.1 are computed as follows, in line with the official metadata, 8 where the numerator includes the number of individuals with at least one of the rights mentioned above: # people in ag. population with ownership or tenure rights over agricultural land 5.a.1 (a) = ×100, by sex Total agricultural population # women in the ag. population with ownership or tenure rights over agricultural land 5.a.1 (b) = ×100, by tenure type Total in the ag. population with ownership or tenure rights over agricultural land We also construct indicator 1.4.2 according to the indicator metadata, which is largely informed by the “Voluntary Guidelines for the Responsible Governance of Tenure of Land, Forests and Fisheries in the Context of National Food Security”. 9 The United Nations World Committee on World Food Security endorsed these guidelines in 2012, and thus they can be viewed as an internationally accepted standard (FAO, 2012). The sub-indicators for 1.4.2 are computed as follows: # adults with legally recognized documentation 1.4.2 (a) = Total adult population ×100, by sex and tenure type # adults who perceive their rights as secure 1.4.2 (b) = ×100, by sex and tenure type Total adult population Sub-indicator 1.4.2 (b) is made up of two components, the perception of tenure security and the right to bequeath, where both are necessary conditions in order for an individual to be considered as having secure rights under the sub-indicator. In the ALTA study, perception of tenure security is asked on a scale from 1 to 5, ranging from “not at all likely [to involuntarily lose land]” to ‘extremely likely [to involuntarily lose land]”, in line with the joint module by FAO, World Bank, and UN-Habitat (2019). The appropriate threshold for what constitutes a secure perception on this scale is not identified in the metadata, leaving room for various interpretations. As this is a subjective dimension, we avoid any assumption about what the appropriate threshold for the secure perception of land would 7 The agricultural population is defined in the SDG 5.a.1 metadata as: “adult individuals living in agricultural households, i.e. households that operated land for agricultural purposes and / or raised livestock over the past 12 months, regardless of the final purpose of the production.” 8 SDG indicator 5.a.1 metadata is available here: https://unstats.un.org/sdgs/metadata?Text=&Goal=&Target=5.a. 9 SDG indicator 1.4.2 metadata is available here: https://unstats.un.org/sdgs/metadata/files/Metadata-01-04-02.pdf 9 be. Rather, in the primary analysis reported in the paper, we make use of the strictest definition, where an individual is considered to have secure land tenure perception if they answer Not at all likely. As a robustness check, we have also replicated the analysis using looser thresholds for secure perception (as reported in Annex Table A1) and the conclusions remain consistent (results not reported).10 V. Results and Discussion The present analysis investigates the differences arising when collecting information on land tenure at different levels of aggregation and type of respondent. What are the expected effects of changing the level of aggregation and the type of respondent on the resulting measure of land tenure security? From a perspective of behavioral science, under conditions of perfect information and no misreporting, one could expect that the indicators built from proxy respondents should be similar to those estimated using self-respondents. However, perfect information is not easily attainable, and recent evidence studying biased beliefs on observable characteristics has shown that even when information is revealed ex-post, bias may persist despite information being readily available (Proto and Sgroi, 2017). Under conditions of imperfect information, there are several dimensions that may affect the likelihood of misreporting by proxy respondents. For example, someone reporting for another individual may have a misconception of his/her level of perceived land tenure security, especially if the reporting and reported individuals do not share the same gender and reside in rural contexts characterized by unequal gender conditions (Namubiru-Mwaura, 2014). With respect to the level of data collection, the level of aggregation (i.e., parcel or aggregate level) may introduce a systematic misreporting of land tenure rights. The direction and magnitude of the error stemming from aggregation is not easily predictable, as it is linked to cognitive processes that respondents employ in aggregating such information. Since the present study aims at measuring the 1.4.2 and 5.a.1 indicators, disentangling the behavioral mechanism falls beyond our scope. These mechanisms, however, are at the core of a second study based on a robust empirical analysis targeting several types of heterogeneities in misreporting (Gourlay et al., 2021). The analysis proceeds by comparing the effect of collecting data at the aggregate level (Arm 2), through proxy respondents (Arm 3), and with both proxy respondents and aggregate-level collection (Arm 4), relative to a self- reported parcel-level approach (Arm 1), which represents the gold standard for these types of surveys. First, we investigate how the different components of land tenure security vary across treatment arms, indicating the effect of the related survey designs. Second, we present the resulting SDG indicators, both for the overall population and by gender. Finally, we examine the gender gap in the land tenure components and SDG indicators, exploring how these gaps vary with different respondent approaches and levels of land aggregation. As the results in Table 3 suggest, the point estimates of the land tenure components, namely legal documentation, right to sell, right to bequeath, and perception of security, vary across the four treatment arms, but these differences are generally not statistically different from the means of Arm 1, with the exception of Arm 3. 11 Data collected 10 The tenure typology does not vary substantially in our data, with more than 95% of parcels reported as private tenure. Tenure type is not available for the modules where land data is collected at the aggregate level. For this reason, we have excluded disaggregation by tenure type from the analysis. 11 Due to an incorrect enabling condition in the CAPI application, the question asking whether a specific person’s name was included on a legally recognized document was skipped for all non-agricultural land in Arm 4. The question of whether a document exists at the household level was asked, but the follow-up question asking about a given person’s name being present on that document was not. This has the potential to bias the Arm 4 results for the estimation of SDG 1.4.2(a) and the incidence of individuals with their name on legal documentation (for the total population only, not for analysis restricted to agricultural land as for 5.a.1). However, given the low incidence of households with members who reported rights to non- 10 through Arm 3, using the proxy respondent, parcel-level questionnaire approach, results in a statistically significant underestimation of the share of individuals reporting the right to sell, the right to bequeath, and a perception of secure tenure (Table 3). 12 For example, when comparing Arm 1 to Arm 3, the share of respondents reporting the right to bequeath land decreases from 40 to 29 percent. The incidence of legal documentation, however, remains statistically stable across all arms, including Arm 3. In general, the observed differences in the incidence of land tenure rights under Arm 3 is particularly important, as the parcel-level proxy respondent approach is commonly employed in agricultural surveys. These same dimensions measured through the questionnaire collecting data at an aggregate level and with self- response (Arm 2) results in point estimates slightly above the ones of Arm 1, but the difference in means is not significant. When using a questionnaire collecting data at an aggregate level but using proxy respondents, point estimates are slightly lower than the gold standard of Arm 1, but again are not statistically different. These observed differences in point estimates may have implications when comparing individual land rights across time or space (e.g., across countries) if the surveys being compared employed different approaches to data collection. Figure 2 illustrates the sub-indicators for 1.4.2 and 5.a.1 by arm alongside their respective 95 percent confidence intervals. In terms of tenure security for the overall population, approximately 60 percent of individuals are considered to have secure tenure rights as measured in Arm 1, according to indicator 1.4.2 (a), which is based on legally recognized documentation. The same sub-indicator decreases when obtained through Arm 3 and Arm 4, while it slightly increases for Arm 2, although the means are not significantly different than those estimated through Arm 1. Indicator 1.4.2 (b), based on perception of land tenure security, shows more variation across arms – Arm 3 estimates 22 percent of individuals perceive their land tenure to be secure, which is significantly different from the Arm 1 estimate of 33 percent of individuals. For 5.a.1 (a), we estimate a higher share of land-secure individuals when comparing the aggregate-level approach (Arms 2 and 4) to the gold standard, and a lower share of land- secure individuals using Arm 3. However, these differences from Arm 1 are not statistically significant. Similarly, there are no statistically significant differences in the estimates of 5.a.1 (b) between Arms 2, 3, 4 and Arm 1. In operational terms, if financial or technical reasons require survey practitioners to select an aggregated approach using proxy respondents, our findings suggest that in the case of Armenia, the errors stemming from aggregation and proxy respondent bias appear to compensate for each other, specifically in the construction of SDG 5.a.1, which takes an ‘either/or’ approach to indicator construction (that is, if an individual has legally recognized documentation or the right to bequeath or the right to sell, they are considered secure). This finding may be specific to contexts like Armenia, where documentation rates are relatively high. The significant trade-offs in terms of the analytical value of parcel-level versus aggregate-level land data must also be carefully considered if opting for the Arm 4 questionnaire. agricultural land only, and the consistency observed between the findings of total land and agricultural land only (which is not impacted by the error), we believe the potential bias to be negligible. At the household level, only 3 households of all 301 in Arm 4 reported rights to non-agricultural land only (including a total of 7 women and 5 men). At the individual level, in total, 21 adults (7 men and 14 women) reported rights to non-agricultural land only, of which 3 (women) reported there was no document for the relevant land, leaving a total of 18 individuals, or 1.5% of the adults sampled in Arm 4, with an uncertain “name on document” status. 12 Table 3 summarizes the various land tenure components of the two indicators by arm for the total sample population, while results for the agricultural population (and agricultural land in particular) are presented in Table A2 of Annex II. 11 Table 3. Summary statistics of land security components by arm and t-test of each arm with respect to Arm 1 Self- Respondent, Self-Respondent, Proxy Respondent, Proxy Respondent, Parcel Level Aggregate Level Parcel Level Aggregate Level (Arm 1) (Arm 2) (Arm 3) (Arm 4) Mean SD Mean SD T-test Mean SD T-test Mean SD T-test Total population Name on document (1=yes) 0.48 0.50 0.50 0.50 0.38 0.47 0.50 0.66 0.47 0.50 0.56 Right to bequeath (1=yes) 0.40 0.49 0.41 0.49 0.63 0.29 0.45 0.00 0.36 0.48 0.12 Secure tenure perception (1=yes) 0.50 0.50 0.54 0.50 0.19 0.38 0.49 0.00 0.50 0.50 0.84 Right to sell (1=yes) 0.40 0.49 0.43 0.49 0.44 0.34 0.47 0.01 0.38 0.48 0.26 Note: The table reports the summary statistics of the land security components by arm with t-tests on the difference with respect to Arm 1, including all land irrespective of use type. Table A2 reports the statistics for agricultural land, for the agricultural population. Figure 1. Sub-indicators 1.4.2 and 5.a.1, by arm Note: The figure displays the value and 95% confidence interval of the sub-indicators 1.4.2 and 5.a.1 on land tenure security. 12 i. Gender gaps and land rights measurement To understand whether the modules differ in terms of their ability to capture gender gaps in land rights, in Table 4 we report the same statistics as in Table 3 above, but disaggregated by gender, and with t-tests across gender within each arm. As Table 4 suggests, the gender gap in land rights is prevalent in Armenia. Independent from the methodology applied, men appear to have higher levels of land tenure security relative to women under both the 1.4.2 and 5.a.1 definitions. The shares of men reporting to have legal documentation, the right to bequeath, the right to sell, and perception of secure tenure, are higher than women, regardless of the arm, and these gender differences are consistently statistically significant. For example, using the gold standard (Arm 1), a wide difference can be observed on the right to sell, where about 53 percent of men report holding this right, as compared to only 31 percent of women. Most notably given the objectives of this paper, for all the dimensions under study, the gap between Arm 1 and the other arms is larger for men than for women, which has significant implications for the estimated gender gap in land rights. The SDG indicators are presented by gender in Figure 2, along with the implied gender gap in rights in Figure 3. 13 Again, the strongest message emerges for the perception of land tenure security, 1.4.2 (b), with the estimates for the share of men with a perception of secure tenure differing most widely across arm. The results suggest that using an Arm 3 approach may result in a statistically significant underestimation of men’s perception of land tenure security relative to the gold standard. In Figure 4, the gender gap is presented by sub-indicator (with 5.a.1 (b) excluded, as it focuses only on women). The point estimates of the gender gap vary compared to the gold standard for each sub-indicator, but the differences between the gold standard and each of the alternative treatment arms are not statistically significant in a t-test. 13Figures 3 and 4 do not display indicator 5.a.1 (b), as this is the share of women among rights holders and thus it cannot be disaggregated by gender. 13 Table 4. Gender-specific statistics for components of land tenure security and t-test across gender within each arm Self-Respondent, Parcel Self-Respondent, Aggregate Proxy Respondent, Parcel Proxy Respondent, Aggregate Level (Arm 1) Level (Arm 2) Level (Arm 3) Level (Arm 4) Variables Male Female T-test Male Female T-test Male Female T-test Male Female T-test Total population Name on the document (1=yes) 0.60 0.39 0.00 0.63 0.40 0.00 0.56 0.38 0.00 0.55 0.39 0.00 Right to bequeath (1=yes) 0.52 0.31 0.00 0.54 0.31 0.00 0.34 0.23 0.00 0.43 0.30 0.00 Secure perception (1=yes) 0.57 0.45 0.00 0.61 0.49 0.00 0.47 0.30 0.00 0.56 0.45 0.00 Right to sell (1=yes) 0.53 0.31 0.00 0.56 0.32 0.00 0.42 0.26 0.00 0.45 0.31 0.00 Note: the table reports the summary statistics of the land security components by arm and gender, with the respective t-tests on the difference between gender, for all land irrespective of use type. Table A3 reports the statistics on agricultural land for the agricultural population. 14 Figure 2. Sub-indicators 1.4.2 (a), 1.4.2 (b), and 5.a.1 (a), by gender and arm Note: the figure displays the value of sub-indicators 1.4.2 (a), 1.4.2 (b) and 5.a.1 (a) by gender and arm, with the distance between the two points indicating the gender gap. Figure 3. Gender gap in the sub-indicators of land rights, by arm 15 ii. Practical implementation implications The implementation of these different modules does not come without a cost. The selection of the appropriate module depends on a set of considerations including, among others, financial constraints of the survey operation, implementation timelines, the structure of the survey in which the module is being integrated, and the associated respondent fatigue, which recent studies have linked to increased bias in rural livelihoods reporting (see, for example, Ambler et al., 2021). Implementing a self-respondent approach does require additional time and coordination to appropriately identify (in a random manner) the selected individual(s) for the interview as well as to schedule the interview at a time that is convenient for the respondent(s). Relatedly, as in the case of Armenia, some household members may move outside the community for work, resulting in a biased sample of individuals who complete the self-respondent interview. This is explored below, following the discussion of interview duration for the various modules.14 Figure 5 presents the distribution of the duration of the land tenure module interviews, by arm. As expected, the interviews based on the parcel-level modules (Arm 1 and Arm 3) are longer than those at the aggregate level. The average time for completion of the Arm 1 module, including all respondents interviewed per household, is about 12.1 minutes, a value not statistically different than that for Arm 3 (10.8 minutes), suggesting few or null advantages in terms of time in implementing a proxy respondent approach over a self-respondent approach if collecting data at the parcel level. This finding should be tempered, however, by the fact that the interview duration only accounts for actual interviewing time, not the time needed to identify and coordinate with each self-respondent. The aggregate-level arms, Arms 2 and 4, are substantially faster than those implemented at the parcel level. Respondents in Arm 2 take on average about 3 minutes to complete the module per household (that is, aggregating all respondent interviews per household), which is lower and significantly different than the Arm 4 mean of approximately 3.6 minutes. 15 Duration of the land tenure module interview .25 .2 Kernel density .1 .05 0 .15 0 20 40 60 80 minutes ARM1 ARM2 ARM3 ARM4 Figure 4. Distribution of the land tenure module interview duration, by arm 14 Given complexities in the survey design and data collected beyond land tenure, including the testing of various approaches to measuring land area, we report differences in costs across modules in terms of land tenure module interview duration only. Monetary estimates per treatment arm would not accurately reflect costs related to the land tenure module alone and would vary significantly with survey context. 15 Note that interview durations are aggregated at the household level and exclude any interviews that spanned multiple days at the module-respondent level, as duration estimates for those interviews are prone to error. Estimates for Arm 1 include the time taken to complete the parcel roster. All durations are winsorized at the 95% level to minimize the impact of outliers. 16 One additional way that survey design may affect data quality is by increasing the non-response rate. In the ALTA study, the non-response rates at the individual level for Arms 1 and 2 are 13% and 12%, respectively, with the difference not statistically different from zero. We also assess whether the level of detail of the questionnaire may influence the willingness for participation in the self-respondent modules, Arms 1 and 2. To do so, we conduct a t-test on a set of dummy indicators capturing whether the individual responded to the questionnaire and reason for non-response, if applicable. Note that the first respondent that is interviewed within a given household is not aware of the length or structure of the questionnaire module. The second and/or third interviewed respondents may have a sense, however, if they are informed by the first respondent. As Table 5 suggests, neither the differences in the share of individuals not responding to Arms 1 and 2, nor the reason for not responding, are significant. The only notable exception is the share of individuals not responding as out of marz, but this does not appear to be linked with the length of the questionnaire itself. Table 5. Non-response rate and reason for non-response in self-respondent arms Self-Respondent, Self-Respondent, Parcel Level Aggregate Level (Arm 1) (Arm 2) Variables Mean Mean T-test Non-response (1=yes) 0.13 0.12 0.43 Non-response as out of the marz (1=yes) 0.08 0.05 0.03 Non-response as ill/not well (1=yes) 0.02 0.02 0.83 Non-response as refused (1=yes) 0.00 0.01 0.17 Other reason for non-response (1=yes) 0.03 0.04 0.28 Table 6 reports the non-response rates by gender to investigate whether there are differences in the non-response rate and the reasons for non-response based on gender. These findings indicate that amongst individuals that are randomly selected for interviewing, there is a higher rate of non-response for men relative to women in both Arms 1 and 2. In terms of magnitude, males appear 3.5 times less likely than women to respond to the self-respondent parcel-level questionnaire, and 1.5 times less likely to respond to the aggregate-level questionnaire. The migration of men out of the marz for periods of time, likely for employment, is a main driver of this increased non-response. Finally, we do not observe any statistical difference in the rate of outright refusal, which may indicate that the difference in the rate of non-response across men and women is driven by circumstantial and socio-economic reasons as opposed to behavioral differences. 17 Table 6. Non-response, by gender of self-respondent Self-Respondent, Self-Respondent, Parcel Level (Arm 1) Aggregate Level (Arm 2) Variables Male Female T-test Male Female T-test Non-response (1=yes) 0.21 0.06 0.00 0.14 0.09 0.03 Non-response as out of the marz (1=yes) 0.13 0.03 0.00 0.07 0.03 0.00 Non-response as ill/not well (1=yes) 0.03 0.01 0.08 0.02 0.02 0.80 Non-response as refused (1=yes) 0.01 0.00 0.49 0.01 0.01 0.34 Other reason for non-response (1=yes) 0.04 0.01 0.04 0.04 0.03 0.45 VI. Conclusion Individual-level data on land rights is essential for understanding how land rights differ by gender, as well as how those differences affect development outcomes. SDG indicators 1.4.2 and 5.a.1 both require data on individual land rights, encouraging countries to relinquish previous practices of collecting household-level land data that implied an assumption of a unitary household model. However, approaches to collecting individual-level land data are varied and different methodological choices may result in biased understandings of the importance and level of land rights by gender. A growing body of literature has shown that survey design decisions on the approach to respondent selection can significantly impact estimates across a wide range of topic areas. This paper contributes to that literature by examining the implications of respondent strategy and the level of land data disaggregation on the computation of SDGs 1.4.2 and 5.a.1 and the distribution of the underlying land rights, by gender. Using data from the Armenia Land Tenure and Area study – a study designed specifically for this analysis – we compare the implications of the use of a proxy respondent versus the recommended self-respondent approach and the use of aggregated land data versus parcel-level land data (recommended). Understanding the implications of these design decisions on data quality is critical, as household surveys generally rely on proxy respondents and may use an aggregate- or parcel-level approach depending on the nature of the survey. In this initial comparative assessment, we find that in the case of Armenia, the measurement of legally documented rights is robust to these design decisions, both for men and women, when compared to the gold standard self-respondent, parcel-level approach. This consistency in the measurement of legal documentation also translates into consistent measurement of SDG indicators 1.4.2 (a) and 5.a.1. Land rights that are less objective in nature and potentially more difficult for proxy respondents to answer, including the right to sell, the right to bequeath, and the perception of tenure security, are more sensitive to the data collection approach. We find that collecting data at the parcel level through a proxy respondent underestimates the share of individuals holding these types of rights to a statistically significant degree. The degree of variability in these rights across treatment arms also differs by gender. The estimates for the share of men with a perception of secure tenure as measured through a proxy, parcel-level approach differ significantly from those measured by the gold standard, while women’s perception of security does not. Reviewing the gender gap between men’s and women’s rights, we find that the point estimates of the gap between the gold standard and the other treatment 18 arms is larger for men than for women, although the resulting gender gaps in SDG indicators are not statistically different. The findings of this paper suggest that SDG indicators 1.4.2 (a) and 5.a.1 can be reasonably monitored through the use of proxy respondents and/or aggregated land data, in the case of Armenia and potentially in other cases with relatively high rates of documented land rights. However, in conducting analyses of land rights beyond SDG monitoring, biases are still observed with the use of these design decisions for rights other than legal documentation, including the right to sell, right to bequeath, and perception of tenure security. In future work, the behavioral mechanisms behind the biases suggested by the proxy respondent and aggregation approaches will be unpacked. Given the uniqueness of the Armenian land context, the study will also be replicated in an alternative context to assess the generalizability of the results to areas with lower rates of documented land rights. 19 VII. References Ambler, K., Herskowitz, S., & Maredia, M. K. (2021). Are we done yet? Response fatigue and rural livelihoods. Journal of Development Economics, 102736. Asfaw, S., & Maggio, G. (2018). Gender, weather shocks and welfare: Evidence from Malawi. The Journal of Development Studies, 54(2), 271-291. Bambio, Y., & Agha, S. B. (2018). Land tenure security and investment: Does strength of land right really matter in rural Burkina Faso?. World Development, 111, 130-147. Carter, M., and Barrett, C. (2006). “The economics of poverty traps and persistent poverty: An asset-based approach.” The Journal of Development Studies, 42.2, pp. 178–199. Beegle, K., De Weerdt, J., Friedman, J., & Gibson, J. (2012). Methods of household consumption measurement through surveys: Experimental results from Tanzania. Journal of Development Economics, 98(1), 3-18. De Mel, S., McKenzie, D. J., & Woodruff, C. (2009). Measuring microenterprise profits: Must we ask how the sausage is made?. Journal of Development Economics, 88(1), 19-31. Deininger, K., Xia, F., Kilic, T., & Moylan, H. (2021). Investment Impacts of Gendered Land Rights in Customary Tenure Systems. World Development, 147. https://doi.org/10.1016/j.worlddev.2021.105654 Domeher, D., & Abdulai, R. (2012). Access to Credit in the Developing World: does land registration matter?. Third World Quarterly, 33(1), 161-175. Doss, C. (2013). Intrahousehold bargaining and resource allocation in developing countries. The World Bank Research Observer, 28(1), 52-78. Doss, C., Kovarik, C., Peterman, A., Quisumbing, A., & Van Den Bold, M. (2015). Gender inequalities in ownership and control of land in Africa: myth and reality. Agricultural Economics, 46(3), 403-434. Food and Agriculture Organization of the United Nations (2012). Voluntary Guidelines on the Responsible Governance of Tenure of Land, Fisheries and Forests in the Context of National Food Security. Available at: http://www.fao.org/docrep/016/i2801e/i2801e.pdf FAO; The World Bank; UN-Habitat. (2019). Measuring Individuals’ Rights to Land: An Integrated Approach to Data Collection for SDG Indicators 1.4.2 and 5.a.1. Washington, DC: World Bank. © FAO, The World Bank, and UN-Habitat. License: CC BY-NC-SA 3.0 IGO Fisher, M., Reimer, J. J., & Carr, E. R. (2010). Who should be interviewed in surveys of household income?. World Development, 38(7), 966-973. Giovarelli, R., & Bledsoe, D. (2001). Land reform in Eastern Europe. Western CIS, Transcaucuses, Balkans, and EU Accession Countries. Hasanbasri, A., Kilic, T., Koolwal, G., & Moylan, H. (2021a). LSMS+ Program in Sub-Saharan Africa: Findings from Individual-Level Data Collection on Labor and Asset Ownership. Washington, D.C.: World Bank Group. Hasanbasri, A., Kilic, T., Koolwal, G., & Moylan, H. (2021b). LSMS+ Program in Cambodia: Findings from Individual-Level Data Collection on Labor and Asset Ownership. Washington, D.C.: World Bank Group. 20 Hasanbasri, A., Kilic, T., Koolwal, G., Moylan, H. (2022). Multidimensionality of Land Ownership among Men and Women in Sub-Saharan Africa. World Bank Policy Research Working Paper No. 9900. The World Bank. Joel, T. K. E., & Bergaly, K. C. (2020). Land Tenure Security, Credit Access and Agricultural Productivity in Cameroon. African Economic Research Consortium, Research Paper 395. Kilic, T., & Moylan, H. (2016). Methodological Experiment on Measuring Asset Ownership from a Gender Perspective (MEXA): Technical Report. Washington, D.C.: World Bank. Kilic, T., Moylan, H., & Koolwal, G. (2020). Getting the (gender-disaggregated) lay of the land: impact of survey respondent selection on measuring land ownership and rights. World Bank Policy Research Working Paper No. 9151. The World Bank. Kilic, T.; G. Van den Broeck; G. Koolwal; and H. Moylan. (2020). Are You Being Asked? Impacts of Respondent Selection on Measuring Employment. World Bank Policy Research Working Paper 9152. The World Bank. Lawry, S., Samii, C., Hall, R., Leopold, A., Hornby, D., & Mtero, F. (2017). The impact of land property rights interventions on investment and agricultural productivity in developing countries: a systematic review. Journal of Development Effectiveness, 9(1), 61-81. Lerman, Z., & Mirzakhanian, A. (2001). Private agriculture in Armenia. Lexington Books. Moore, J. C. (1988). Self/proxy response status and survey response quality. Journal of Official Statistics, 4(2), 155-172. Namubiru-Mwaura, E. (2014). Land Tenure and Gender : Approaches and Challenges for Strengthening Rural Women's Land Rights. World Bank, Washington, DC. © World Bank. Proto, E., & Sgroi, D. (2017). Biased beliefs and imperfect information. Journal of Economic Behavior & Organization, 136, 186-202. Statistical Committee of the Republic of Armenia. (2016). Main Findings of Agricultural Census 2014 of the Republic of Armenia. http://armstat.am/en/?nid=82&id=1860 Statistical Committee of the Republic of Armenia. (2021). Statistical Yearbook of Armenia. https://armstat.am/file/doc/99526883.pdf Todorov, A., & Kirchner, C. (2000). Bias in proxies' reports of disability: data from the National Health Interview Survey on disability. American Journal of Public Health, 90(8), 1248. UNECE. (2005). Inventory of Land Administration Systems in Europe and North America. Produced and published by HM Land Registry. London, United Kingdom on behalf of the UNECE Working Party on Land Administration. United Nations Statistics Division (UNSD). (2019). Guidelines for Producing Statistics on Asset Ownership from a Gender Perspective. Retrieved from: https://unstats.un.org/edge/publications/docs/Guidelines_final.pdf Zhang, L., Cheng, W., Cheng, E., & Wu, B. (2020). Does land titling improve credit access? Quasi-experimental evidence from rural China. Applied Economics, 52(2), 227-241. 21 Annex I. Description of Unified Land Tenure Questionnaire Modules 16 Due to the various survey designs in which this module could be integrated, multiple versions of the module have been designed to facilitate its inclusion in a number of different survey designs and types. The appropriate version selection for a given survey depends on: (i) whether one person responds on behalf of other household members (proxy respondent approach) or each individual is asked specifically about his or her land tenure rights (self- respondent approach); and (ii) whether data is collected at the parcel level or the household/farm level. Five versions of the questionnaire module are provided, to account for the various combinations of respondent selection and level of data collection (refer to the attached questionnaire modules and guidance note for more detail). The versions are differentiated as follows: Version 1: Parcel-level data; self-respondent approach; no parcel roster elsewhere; assumes separate household member roster which records individual sex; administered to (a) one or more randomly selected individuals or (b) all adult household members. Version 2: Parcel-level data; self-respondent approach; assumes parcel roster elsewhere which can be fed forward into either (a) the interview of one or more randomly selected individuals or (b) the interviews of all adult household members; assumes separate household member roster with sex. Version 3: Individual-level data; self-respondent approach; not reported at parcel level; administered to (a) one or more randomly selected individuals or (b) all adult household members. Version 4: Parcel-level data; proxy respondent approach; no parcel roster elsewhere; assumes separate household member roster with sex. Version 5: Individual-level data; proxy respondent approach; not reported at parcel level. Each version offers advantages and disadvantages. The use of proxy respondents has the potential to result in biased results but can simplify fieldwork operations and reduce fieldwork time. Similarly, collecting data at the parcel level will produce the most valuable, disaggregated data on land tenure, but may be costlier than asking questions at the individual or farm level. 16 This annex is an excerpt from FAO, World Bank, and UN-Habitat (2019). 22 Annex II. Annexed Tables Table A1. Defining land tenure perception How likely is [PARCEL/LAND] to be taken away from you against your will in the next 5 years? Primary Analysis Robustness #1 Robustness #2 Not at all likely Secure Secure Slightly likely Secure Moderately likely Not secure Very likely Not secure Not secure Extremely likely Table A2. Summary statistics of land security components by arm for the agricultural population Self- Respondent, Self-Respondent, Proxy Respondent, Proxy Respondent, Parcel Level Aggregate Level Parcel Level Aggregate Level (Arm 1) (Arm 2) (Arm 3) (Arm 4) T- Mean SD Mean SD test Mean SD T-test Mean SD T-test Agricultural land Name on document (1=yes) 0.59 0.49 0.63 0.48 0.17 0.58 0.49 0.77 0.60 0.49 0.59 Right to bequeath (1=yes) 0.49 0.50 0.51 0.50 0.55 0.35 0.48 0.00 0.46 0.50 0.34 Secure tenure perception (1=yes) 0.61 0.49 0.68 0.47 0.03 0.48 0.50 0.00 0.65 0.48 0.25 Right to sell (1=yes) 0.50 0.50 0.53 0.50 0.34 0.42 0.49 0.01 0.48 0.50 0.58 Note: The table reports the summary statistics of the land security components by arm with t-tests on the difference with respect to Arm 1 for agricultural land, for the agricultural population. 23 Table A3. Summary statistics of land security components by gender and arm for the agricultural population Self-Respondent, Proxy Respondent, Proxy Respondent, Self-Respondent, Parcel Aggregate Level Aggregate Level Parcel Level (Arm 3) Level (Arm 1) (Arm 2) (Arm 4) T- T- T- T- Variables Male Female Male Female Male Female Male Female test test test test Agricultural land Name on the 0.72 0.48 0.00 0.78 0.51 0.00 0.67 0.49 0.00 0.69 0.52 0.00 document (1=yes) Right to bequeath 0.62 0.39 0.00 0.65 0.38 0.00 0.41 0.29 0.00 0.53 0.39 0.00 (1=yes) Secure perception 0.68 0.56 0.01 0.75 0.62 0.00 0.57 0.38 0.00 0.70 0.59 0.00 (1=yes) Right to sell 0.64 0.38 0.00 0.68 0.40 0.00 0.51 0.34 0.00 0.55 0.41 0.00 (1=yes) 24