Policy Research Working Paper 10847 Connectivity, Road Quality, and Jobs Evidence from Armenia Nino Pkhikidze Transport Global Practice July 2024 Policy Research Working Paper 10847 Abstract Good road infrastructure decreases travel time and improves the Caucasus from 1903. The results show that a shorter accessibility to urban areas. Improved rural-urban link- distance to a good quality road has a statistically significant ages could also affect rural employment through decreased positive impact on overall non-agricultural employment for time and travel costs. To study this link, the paper ana- men and women, increasing the likelihood of cash-earning lyzes the impact of good quality roads on agricultural and jobs for rural women and skilled manual and non-seasonal non-agricultural jobs in Armenia, using different sets of employment for rural men. People are more likely to work data and different methodological approaches. To address outside their villages and work for more hours if they have endogeneity and reverse causality issues of road quality, access to good quality roads. The results are robust from the the paper uses a historical instrumental variable obtained analysis of Demographic and Health Survey as well as the by digitizing historical roads which were mainly used for Integrated Living Conditions Survey of Armenia. military purposes—from a military-topographic map of This paper is a product of the Transport Global Practice. 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 author may be contacted at npkhikidze@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 Connectivity, Road Quality, and Jobs: Evidence from Armenia Nino Pkhikidze∗ Keywords: Roads, infrastructure, rural employment, historical IV. JEL classification: H54, J60, C26, Q10, R40. ∗ World Bank, 1818 H St. NW, Washington DC, 20433. npkhkidze@worldbank.org. I would like to thank Prof. Joachim von Braun and Alisher Mirzabaev for their guidance on this work. Gilles Duranton, Chiara Kofol, Lukas Kornher, and the participants of different seminars and workshops at the Center for Development Research (ZEF), the University of Bonn, Caucasus Resource Research Center (CRRC) Armenia, and the Development Impact Group (DIME) at the World Bank for their useful comments on the previous versions of this paper. I would also like to thank World Bank’s Armenia Country Office for their guidance and support. The findings, interpretations, and conclusions expressed in this paper are entirely of the author. 1 Introduction Access to jobs, particularly to better jobs, is crucial for economic development. Prox- imity to urban areas has always been considered an important aspect for rural households for accessing jobs, goods, and services, and it is widely studied in the literature. The studies show that living closer to urban areas increases non-farm employment and mar- ket oriented activities (Sheng et al., 2022; Deichmann et al., 2009; Fafchamps and Shilpi, 2005, 2003), improves the economic well-being and nutrition of rural households (Stifel and Minten, 2017; Sharma, 2016), and overall, positively affects spatial dimensions of develop- ment (Sharma, 2016). However, proximity alone cannot be a good measure of accessibility because it does not take into account road condition - which can have a large impact on travel time and costs to urban centers. Poor transportation infrastructure restricts accessibility to markets and jobs for the rural population in low- and middle-income countries (World Bank, 2009). Improved roads tend to decrease travel time and travel costs, and promote mobility. Better transportation infrastructure might stimulate mobility of goods and labor, connecting people to jobs. A large majority of households in low- and middle-income countries are still mainly employed in agriculture, often in self-subsistence farming. According to the accounts data, the average agricultural productivity gap is 3.5, meaning that value added per worker in agriculture is more than three times lower than the one of the non-agricultural sector. There are higher disparities in different income levels of the countries. In low-income countries labor productivity in non-agricultural employment is 4.5 times higher than in agricultural employment, it is 3.2 times higher in middle-income countries, and 2.2 higher in high-income countries (Gollin, 2023; Gollin et al., 2014). Structural transformation has been a key policy issue in low- and middle-income countries in recent decades. One of the ways to address this problem is to give rural households opportunities to be employed in more productive sectors. Proximity to cities and access to adequate road infrastructure could play an important role in increasing non-agricultural employment among rural people (Aggarwal, 2018; Asher and Novosad, 2020). This paper analyzes the impact of road quality on agricultural and non-agricultural employment, studying the evidence from Armenia. Empirical studies on the relationship 2 between road infrastructure and labor outcomes often suffer from endogeneity and reverse causation problems. Road quality is usually not exogenous. Other than topographic characteristics, it is often driven by economic, social or political factors (Banerjee et al., 2012; Datta, 2012; Nguyen et al., 2011; Burgess et al., 2015). To address the endogeneity and reverse causation issues, this study uses a (historical) instrumental variable strategy. The instrumental variable has been obtained by georeferencing and digitizing a military- topographic map of the Caucasus region printed by the Russian Empire in 1903. The study uses historical, primary military and post roads to instrument the existing road quality a century later. The results show that a shorter distance to the nearest good quality road has statisti- cally significant positive impact on overall non-agricultural employment, on skilled manual employment for rural men, and non-agricultural employment and cash earnings for rural women. People are more likely to work outside of their villages if they have access to good quality roads, and also tend to work for more hours. The analysis has been carried out on two different datasets, the Demographic and Health Survey and the Integrated Living Conditions Survey of Armenia, using different estimation methods. The results are similar and robust from both datasets. This paper contributes to three groups of literature: (1) estimating impact of road infrastructure improvement, (2) examining rural employment and structural transforma- tion, and (3) using historical setting for causal inference. There is a growing number of studies evaluating road construction or improvement programs in various countries, like in India (Aggarwal, 2018; Asher and Novosad, 2020; Bell and van Dillen, 2014; Duranton et al., 2014; Datta, 2012; Ghani et al., 2016; Herrera Dappe et al., 2021), Bangladesh (Khandker et al., 2009; Khandker and Koolwal, 2011), Papua New Guinea (Gibson and Rozelle, 2003; Wiegand et al., 2017), Ethiopia (Dercon et al., 2009), Indonesia (Gibson and Olivia, 2010; Gertler et al., 2022), Viet Nam (Mu and van de Walle, 2011), China (Banerjee et al., 2012; Wang et al., 2016; Fan and Chan-Kang, 2005; Faber, 2014), Geor- gia (Lokshin and Yemtsov, 2005), among others. Overall, improved connectivity reduces travel time and cost. The cost and time saving stimulates mobility, connecting rural areas to urban centers, people to markets and services. Easier mobility quite often is a key factor in trading, as a result reducing product prices (Donaldson, 2018; Andrabi and Kuehlwein, 3 2010; Aggarwal, 2018), it also reduces poverty (Khandker et al., 2009), and contributes in local market development (Mu and van de Walle, 2011) and welfare gains (Kebede, 2024).1 The second area of contribution is to the literature on rural employment and structural transformation. Road improvement and travel time reduction contributes in easing access to jobs. High transportation costs lead to increased size of the agricultural workforce and employment in subsistence farming (Gollin and Rogerson, 2014; Adamopoulos, 2011). On the other hand, improved accessibility to jobs, through improved roads, benefits structural transformation. Asher and Novosad (2016) and Mu and van de Walle (2011) find that the rural roads programs in India and Viet Nam, respectively, increase wage labor participation and the share of households mainly relying on the service sector as their main source of income. Nakamura et al. (2020) show that improvement of rural roads in Ethiopia led to a 4.6 percentage point increase in women’s employment and 14.1 percentage point increase in youth employment. In India, Herrera Dappe et al. (2021) found that the construction of new all-weather rural roads resulted in a 9.5 percent increase in overall employment rates. This effect was more pronounced in villages located farther from urban centers. Improved rural roads have also been shown to increase household income by specializing in agriculture by using more fertilizer and newer technologies, and hiring more labor, as shown by Qin and Zhang (2016) and Shamdasani (2021) in the cases of Chinese and Indian rural roads, respectively. And finally, the paper also contributes to the growing body of literature that use historical settings to account for endogeneity. Duranton and Turner (2012) use a historical highway plan of the US highway system to estimate a structural model of city growth and transportation, Baum-Snow et al. (2012) use Chinese rail and road networks from 1962 as a source of identifying variation in rail and road networks after 2000, Volpe Martincus opez et al. (2015) and Holl (2016) et al. (2017) use historical Inca routes for Peru, Garcia-L´ use Roman roads and the 1760 Bourbon postal routes as sources of exogenous variation of oller and Zierer (2018) rely on an 1890 plan of railroad highway extension in Spain, and M¨ network in Germany and 1937 map of planned autobahns to study regional employment. 1 For an overview on the impacts of infrastructure improvements on various economic outcomes, please see Redding and Turner (2015). 4 This paper uses the historical military routes of Armenia during the Russian Empire times to account for the endogeneity of modern day road quality. The research focuses on Armenia to fill the gap in the literature for several reasons. In Armenia the agricultural employment rate is still high, has reduced from 40.4% in 1991 to only 33.6% in 2016.2 Moreover, most people engaged in agriculture in Armenia, like in many other low- and middle-income countries, are best characterized as engaging in subsistence or quasi-subsistence agriculture, meaning that they consume most of the goods they produce. Given this, identifying mechanisms of structural change is very important. In addition, Armenia, has been receiving funding for rehabilitating and improving existing deteriorated roads, as well as for building new ones. Therefore, it is policy relevant to study the impacts of these road projects. And lastly, the unique dataset on road quality and the historical setting of Armenia give an opportunity to identify impacts of road quality on rural employment outcomes. 2 Conceptual framework In regional research, the importance of transportation costs has long been recognized unen’s land rent model as one of the major factors of economic development. Von Th¨ and its subsequent modifications predict concentric circles of specialization in agricul- unen, 1826). Building on this, regional economic models ture surrounding cities (Von Th¨ predict that improving rural connectivity would facilitate moving from agricultural to non- agricultural employment. For example, the Alonso-Muth-Mills model predicts an urban perimeter beyond which agriculture would be the primary employment sector. The urban perimeter takes into account that urban wages, deduced with transportation costs, would be lower than agricultural wages (Brueckner, 1987). Therefore, if we consider road quality improvements as a source of decreasing commuting costs in a given location, we expect improved road quality to expand the urban perimeter. Starting from Lewis (1954) model, researchers have argued that labor market im- perfections prevent people employed in the agricultural sector from relocating towards 2 Source: International Labor Organization (ILO), Key Indicators of the Labor Market https://www. ilo.org/ilostat. Accessed April, 2019. Note: The recently updated ILO methodology somewhat differs with the one of Armenia’s Statistical Committee. 5 higher productivity sectors.3 Since labor productivity is 4.5 and 3.2 times higher in non- agricultural employment than in agriculture in low- and middle-income countries, respec- tively (Gollin, 2023; Gollin et al., 2014), it is important to study the transition channels from agricultural to non-agricultural employment. A multi-sector multi-region model de- veloped by Gollin and Rogerson (2014) shows that higher transport costs increase the size of agricultural workforce and self-subsistence farming. Deichmann et al. (2009) studies rural-urban linkages in Bangladesh and finds that households living closer to urban cen- ters are more likely to be employed or self-employed in non-farm sector, and Asher and Novosad (2020) find that newly paved rural roads in India increase non-farm employment, however slightly, by stimulating easier access to outside-village labor markets. There are two channels linking road infrastructure and rural employment. The first channel is through agricultural productivity. Improved roads decreasing transportation costs might decrease the costs of agricultural inputs, as shown in (Aggarwal, 2018), help individuals to move to non-agricultural employment by increasing agricultural produc- tivity. For example, Sotelo (2015) estimates on average a 14% increase in agricultural productivity because of paving existing roads, by increasing access to inputs and increas- ing output prices. Shamdasani (2021) shows that rural roads program in India increased the mobility of agricultural workers by the integration of village labor markets across space. The second channel works through decreased job search and commuting costs. De- creased transportation costs likely decrease costs of job search or commuting to a job, reducing barriers to working outside the village. It is an important issue, considering pos- sible underemployment in the agricultural sector. The studies show that agricultural work is less productive than non-agricultural work. However, by looking at actual reported hours worked in different sectors in several Sub-Saharan African countries, McCullough (2017) has shown that the productivity gap is not as high as it was previously considered. She observes that agricultural workers are working far fewer hours than non-agricultural workers, and therefore they can be considered underemployed. The underemployment might be due to the seasonality of agricultural work, as some seasons require far more 3 The literature has also suggested that barriers to the reallocation of labor could result from insurance networks that discourage movement out of rural areas (Munshi and Rosenzweig, 2016), and informational frictions (Banerjee and Newman, 1998), among others. 6 work than others. On the other hand, if farmers have limited access to farm inputs, they might need to work more hours in agricultural sector. Improved accessibility to jobs could help farmers to take non-agricultural work dur- ing the agricultural off-season. Commuting might be particularly hard not only due to long distance to urban areas but also due to poor road quality and poor transportation infrastructure. Improved roads and transportation infrastructure, therefore, are expected to increase the probability of household members engaging in work outside their village. 3 Background The paper focuses on Armenia to study the relationship between road quality and rural employment. Armenia is a lower-middle-income country (GDP per capita USD 3,917 in 2016 (constant 2010 US$)) with a population of 2.9 million people.4 The country has quite a well-developed road network, Figure 1 shows that most of the settlements in the country are connected to roads. However, the quality of roads is still a matter of concern. The railroad network is not very well developed in the country mainly due to the rough terrain. Therefore, road network is vital for passenger and freight transportation (World Bank, 2017). The classified road network of Armenia is 7,700 km long, from which around 1,400 km are interstate roads, 2,520 km are regional roads, and 3,780 km are local roads. Most of the roads were built in 1960s and 1970s, and have deteriorated since independence in 1991 due to poor maintenance (World Bank, 2017). Since the early 2000s the country has received funding from different international organizations, such as the Millennium Challenge Corporation (MCC), the World Bank (WB), and the Asian Development Bank (ADB) to rehabilitate the lifeline roads5 in the country and expand the major interstate road into a highway, connecting the south to the north of the country. Lifeline roads net- work serves as a vital link in the country, connecting 960 communities and approximately 1.1 million rural residents - 37 percent of the total population - with one another and 4 Source: Databank, the World Bank https://data.worldbank.org/indicator Accessed: June 2019. 5 The roads connecting villages to major roads are often called “lifeline roads” in Armenia (World Bank, 2017). 7 with urban centers throughout the country (Mathematica, 2015). There have been several ongoing projects since then: the Rural Roads Rehabilitation Project, which was started by the MCC and taken over and expanded by the World Bank as Lifeline Road Network Improvement Project (LRNIP), and two projects by the ADB: North-South Highway and Rural Roads Sector Project. The projects have been improving road quality and rural- urban connectivity. The Lifeline Road Improvement Project was financed by the World Bank and initiated in 2009. It involved the rehabilitation of 446 kilometers of lifeline roads, approximately 11 percent of the entire Lifeline Road Network of the country. A large share of Armenians are employed in agriculture. Structural transformation has been slow: as of 2016, 33.6% of Armenians were working in agriculture, just 6.8 percentage points lower than in 1991.6 Agriculture accounted for 16.4% of GDP in 2016.7 Like in most low- and middle-income countries, the majority of people employed in agriculture in Armenia are rural individuals employed in self-subsistence or semi-self-subsistence farming. Therefore, it is crucial to understand the link between improved road quality and rural employment, and what role improved roads could play in structural transformation. Finally, the historical setting and available information for Armenia makes it possible to employ instrumental variable strategy to study the causal impacts of road quality. In particular, the paper uses historical roads from the times of the Russian Empire as a conditionally exogenous source of variation in quality of transport infrastructure. 6 Source: International Labor Organization (ILO), Key Indicators of the Labor Market https://www. ilo.org/ilostat. Accessed April 2019. 7 Source: Statistical Committee of the Republic of Armenia, ArmStatBank http://armstatbank.am/ pxweb/hy/ArmStatBank/?rxid=602c2fcf-531f-4ed9-b9ad-42a1c546a1b6 Accessed April 2019. 8 Figure 1: Roads and settlements. Source: Author’s compilation based on WB data on roads, and administrative data from Acopian Center for the Environment. 4 Data In order to estimate the economic impacts of road quality, it is necessary to construct unique settlement-level data combining aggregate and micro-data from multiple sources. This paper uses household and individual level surveys, road quality data, administrative and geospatial data. This section describes the data sources and some summary statistics.8 4.1 Integrated Living Conditions Survey Integrated Living Conditions Survey (ILCS) is a nation-wide survey conducted annu- ally by the National Statistical Service of the Republic of Armenia (Armstat). The survey is representative at country, village/town and marz levels.9 The survey includes rural and urban households and monitors the living standards of households. The questionnaires 8 For more detailed summary statistics please see the Appendix. 9 Source: Quality declaration Integrated Living Conditions Survey of Households, Armstat. https: //www.armstat.am/file/Qualitydec/eng/11.1.pdf Accessed April, 2019. 9 are asked on household as well as the individual level. The survey has been conducted since 2001 onwards, however this study uses data starting 2007 since the questions on in- dividual employment were not asked before the 2007 survey. The repeated cross-sectional data have been adjusted and appended for this study for each year from 2007 to 2016. The main strengths of the data are rich details on individual employment, long time span, and questions about the perception on road and transport quality. However, the main disadvantage of the survey for this study is that no approximate location of house- holds is known, except for the region (Marz) where the household lives and whether the location is rural or urban. The survey contains questions on distances to markets, hospi- tals, banks, and other service centers since 2009, making it possible to control for these characteristics while estimating the impact of road quality. Figure 2: Road quality reported by HHs in 2007-2015 The data on road quality has been collected by asking households living in rural areas how would they rate the quality of roads, from “Poor” condition to “Excellent” condition. The respondents had to evaluate the quality of roads within their settlement or community and roads to regional towns or markets. Figure 2 shows road quality reported by the households over the years of 2007 to 2016. While roads that lead to towns and markets seem to have improved over time, internal village roads remained in poor quality. 4.2 Demographic and Health Survey (DHS) The Armenia Demographic and Health Survey (DHS) 2015-2016 is a nationally repre- sentative sample survey, designed to provide information on population and health issues 10 in Armenia. The data has been collected by the National Statistical Service and the Min- istry of Health of the Republic of Armenia and is co-funded by the United States Agency for International Development (USAID). The main goal of the survey is to collect demo- graphic and health indicators, particularly from women of reproductive age, and in some questions from men as well. In total, 2015-2016 survey has collected data on 6,116 women and 2,755 men of the age range of 15 to 49. The data is representative on the national and rural-urban areas. The 2015-2016 survey has a total of 313 clusters, of which 121 are rural. One of the advantages of the DHS survey is that the program collects the GPS location data of surveyed clusters. In order to protect the confidentiality of the respondents, the locations have been displaced. Each urban cluster has been displaced from the actual location up to 2 kilometers, and each rural cluster up to 10 kilometers.10 The survey also provides a wide range of geospatial covariates, very useful for the purpose of the study. Occupation Men Women Total Not working 27% 62% 50% Professional/technical 9% 25% 17% Clerical 1% 2% 1% Sales 6% 9% 8% Agricultural 26% 45% 36% Services 16% 6% 11% Skilled manual 30% 6% 18% Unskilled manual 9% 7% 8% Other 4% 0% 2% Total 1,233 2,571 3,804 Working 904 981 1,885 Table 1: Rural employment by gender (DHS) Rural respondents report lower labor activity. 43% of DHS respondents report living in rural areas, of which half indicate that they have not worked in last 12 months. The difference is very high between men and women. More than double the women report not working than men - 62% and 27% respectively.11 Table 1 reports rural employment indicators. From those working, women are overwhelmingly employed in agriculture, 45% 10 For more information please see Mayala et al. (2018). 11 Unfortunately, limited employment variables do not fully allow the calculation of the proportion of unemployed people versus the ones who do not work for other reasons. 11 compared to 26% of men, and in professional employment, 25% compared to 9% of men. The employment groups are more varied for men, majority of them reported working in skilled manual employment, 30% compared to only 6% of women. 4.3 Road quality data As in most of the low- and middle-income countries, there was no comprehensive road network quality data available for Armenia. To fill this gap, the World Bank financed the data collection of geo-referenced data on the Armenian road network, road quality, surface type and category. The data were collected in early 2017 using an application RoadLabPro - created by the World Bank as a data collection tool with accelerometers to map a road network, evaluate road condition, detect major road bumps, travel speed and report road safety hazards.12 Figure 3: Road quality by road segment Road quality categories: Very good (IRI range between 1.0 and 2.0), Good (2.0 - 4.0), Fair (4.0 - 6.0), Poor (6.0 - 10.0), Very poor (10.0 - 16.0). Source: Author’s compilation based on data on road quality from the World Bank, and administrative data from the Acopian Center for the Environment. In total, the data were collected on 8,286 km of roads, including major (international and national) and feeder roads. The road quality was divided in five categories according to 12 For more information please see World Bank (2017) p.56. 12 the International Roughness Index (IRI): very good, good, fair, poor, and very poor.13 The data shows that the road quality substantially differs between different regions (Marzes). For example, as Figure 8 shows, Yerevan - the region of the capital city has only 4% of its roads in poor or very poor quality, the reported road quality is better in in Kotayk and Ararat Marzes - surrounding Yerevan - more than 85% of primary and feeder roads are in good or very good condition, while Lori and Shirak regions have only around one-fifth of the primary and feeder roads in good condition. 5 Methodology 5.1 Empirical strategy The impacts of infrastructure are often challenging to capture. This paper studies the questions using several datasets and estimation methods. The first method studies the association between regional road quality and different employment outcomes. It uses the Integrated Living Conditions Survey - 10 years of repeated cross-sectional data - to answer the research questions. The survey has collected rich data on various employment outcomes, which is not usually the case in other surveys, making it possible to look at more variables of interest.14 The first part of the analysis studies the association of reported road quality leading to towns and markets, and binary employment outcomes: whether a person is employed in non-agricultural sector, in seasonal employment, and whether at least one member of the household works outside the village. The second part of the analysis studies the relationship between hours worked, agricultural employment, and road quality. The second method utilizes the opportunity of GPS location information collected by the surveys to answer the research questions. It combines the georeferenced data on quality of roads in Armenia collected by the World Bank, matching the geographic coordinates of household cluster locations collected by the DHS. The combined dataset with geographic locations provides an opportunity to calculate the proximity to good quality roads from 13 Road quality categories: Very good (IRI range between 1.0 and 2.0), Good (2.0 - 4.0), Fair (4.0 - 6.0), Poor (6.0 - 10.0), Very poor (10.0 - 16.0). 14 The questionnaires were slightly modified over the period of 10 years, therefore, the variables have been adjusted accordingly to ensure comparability over the years. 13 each cluster where the household lives, and then estimate the impact of the distance on various employment outcomes. In addition, the availability of georeferenced information allows for the possibility to control for additional geospatial variables which could also be influencing employment outcomes. The idea of quality of roads and employment proba- bility is captured in a simple reduced form specification: P r(Yi = 1|xis ) = ϕ(α + βLogDistGRs + ςIis + δHis + τ Ss + µRs + εis ) (1) The outcome variable Yi of an individual i living in settlement s shows the probability of being employed in the non-agricultural sector, skilled manual employment, seasonal employment, or likelihood of getting cash earnings. The main independent variable is LogDistGRs - log distance to the nearest good/very good quality road. The additional controls include individual level characteristics Iis , household level characteristics His , a vector of settlement (cluster) level geospatial covariates Ss , and region fixed-effects Rs . 5.2 Identification strategy The question of endogeneity usually arises while studying the economic impacts of infrastructure. Initial conditions are likely to determine where are new roads be built or which old roads receive maintenance (Banerjee et al., 2012; Datta, 2012). Roads might be built and kept in good condition in areas with high economic potential. Reverse causality might also be in play: areas with non-agricultural employment potential might require better roads, or better roads might lead to higher non-agricultural employment. So the question could be summed up as follows: do areas with better roads show better non- agricultural employment outcomes because of good road quality, or are better roads at- tracted by (potentially) higher non-agricultural employment? For example, if areas with good quality roads show higher non-agricultural employment, it could be because the roads were built and well-maintained in areas with higher economic potential. Therefore, simple correlation between road quality and non-agricultural employment would overestimate the impact of road quality. Another concern on road quality placement is political or ethnic favoritism. Burgess et al. (2015) show evidence of mis-targeted infrastructure projects in Kenya due to ethnic 14 favoritism, Nguyen et al. (2011) show nepotism of public officials in their communes in Viet Nam, mis-targeting infrastructural projects. Studying the long time span from 1960 to 2010, Jedwab and Storeygard (2017) show that cities around the leader’s place of origin in Sub-Saharan Africa were growing faster than other cities because the leaders favored their cities of origin to target road infrastructure projects. Hodler and Raschky (2014) measured regional favoritism from outer space: they found that subnational regions, where current political leaders were born, have more intense nighttime lights. This paper uses an instrumental variable (IV) strategy to account for endogeneity.15 The IV is based on a map of historical road networks of Armenia obtained from a Military- topographic map of the Caucasus region - shown on Figure 9 - prepared under the Russian Empire in 1903. The argument of exogeneity of the historical setting of roads can be motivated by several reasons. During the beginning of the 20th century Armenia was under the rule of the Russian Empire. The southernmost state of the empire, situated on the border of Ottoman and Persian empires, Armenia was an important territory for military defense. The roads maintained by the Russian Empire were mainly used to transport armies. Even though the roads could be also used for trade and economic reasons, we can argue that the government of the Russian Empire would invest in building, rehabilitating and maintaining the roads necessary for military reasons. The second argument is that, since the historical roads were mapped before the industrialization of the region, when the large majority of people were employed in agriculture, we can argue that the roads would not have been built and maintained to promote non-agricultural employment. It is very unlikely that any decisions made in different settings far back in history, motivated mainly by non-economic reasons, could have anticipated rural employment development a century later.16 15 The literature using similar method includes: Duranton and Turner (2012), Baum-Snow et al. (2012), Volpe Martincus et al. (2017), Garcia-L´ opez et al. (2015), Holl (2016) and M¨oller and Zierer (2018), among others. 16 In the section of robustness checks the paper also addresses the factor of mine locations in the region, and shows that it is highly unlikely that mines would have influenced the historical primary road placement. 15 Figure 4: Digitized Military-topographic map of the Caucasus, 1903, fragment on Armenia. Source: Author’s compilation based on data from the Russian State Library, the World Bank, the Acopian Center for the Environment. Digitization by the author. The historical map was georeferenced and the fragment of the map, containing Arme- nia, was digitized. The original map on Figure 9 shows four types of roads: primary roads, post roads (well developed cart road), cart roads, and drover’s roads. Since drover’s roads are mainly pathways, and would not have been actively used for transporting armies and maintained by the government, they were excluded from the analysis. Figure 4 shows geo- referenced and digitized road map of Armenia. The red lines depict the primary military and post roads of Armenia in 1903. For the instrumental variable Probit regression logarithmic distance to good roads is instrumented by logarithmic distance to primary military and post roads in 1903. LogDistGRs = γLogDistRoads1903s + τ Ss + δHis + µRs + εis (2) P r(Yi = 1|xis ) = Φ(α + βLogDistGRs + ιIis + δHis + τ Ss + µRs + εis ) (3) The instrumental variable LogDistRoads1903s is assumed to be correlated with the 16 endogenous regressor LogDistGRs but independent from the error εs . We can argue that old main roads would have been maintained to transport armies, and later on these roads would also have higher quality. 6 Results The results of the two separate datasets and methods are summarized in different subsections below. 6.1 ILCS First, I show the results from the analysis of the repeated cross-sectional data of Integrated Living Conditions Survey (ILCS). The survey has collected information on perception of respondents on road quality in the region and locally - within the village. The road quality indicators include “good”, “average”, and “poor” (“good” is a reference category in the following results tables). Table 2 shows the marginal effects of the Probit regression. It shows the probability of a respondent being employed in non-agricultural sector (columns (1) and (2)), doing seasonal work (columns (3) and (4)), and whether there is anyone from the household working outside of village (column (5)). Since the distance variables were not included in the questionnaires in 2007 and 2008, the regression is done with and without the distance variables. The first column shows a clear negative association between the household living in a region with average or poor road quality and non-agricultural employment. Individuals reporting poor road quality are 5.4 percentage points less likely to be working in non-agricultural sector than the individuals reporting good quality roads in their regions. Controlling for the distance to the nearest market, kindergarten, and health center lowers the marginal effects, but still holds statistically significant 3.5 percentage points lower probability of working in non-agricultural sector (column (2)). As expected, average roads show lower marginal effects; individuals reporting quality of roads in their region leading to towns and markets as average, are 1.9 percentage points less likely to work in non- agricultural sector than the individuals reporting good quality roads in their regions. The marginal effect changes slightly to 1.4 percentage points (column (2)) when distance 17 variables are included (hence, excluding 2007-2008 datasets because of the lack of the distance variables). Columns (3) and (4) report the probability of working in seasonal employment given different regional road quality leading to towns and markets. Respondents who report average and poor road quality in their regions are, respectively, 1,5 and 5.2 percentage points less likely to work in seasonal employment. Considering that 66% of the respondents employed in agriculture report working whole year round and only 1,3% of them or 370 people report having a second job, this could mean that people living in areas with poor roads are less likely to take seasonal jobs during low agricultural seasons. Lastly, column (5) shows that households are less likely to report any household member working outside of the village if they report having poor or average road quality leading to towns and markets.17 Table 3 reports the relationship between hours worked, agricultural employment, and reported road quality. The first column shows that people who work in agriculture tend to work less, estimated at 16.5 less hours worked the week before the survey, holding all other variables constant and controlling for the survey month. Descriptive statistics show that non-agricultural workers had worked on average 23 hours the week before the interview, while the non-agricultural workers had worked on average 41 hours. This result is in line with the study done by McCullough (2017). By analyzing the Living Standards Measure- ment Study Integrated Surveys on Agriculture (LSMS-ISA) for four African countries - Ethiopia, Malawi, Tanzania and Uganda - she found that each agricultural worker was working on average 700 hours per year compared to the 1850 hours per non-agricultural worker. Our estimates are higher than those of McCullough (2017) but still show signif- icantly large differences between the working hours of agricultural and non-agricultural workers. Column (2) shows that individuals reporting average or poor quality of roads leading to towns and markets, work slightly less than people reporting having good re- gional road quality. Column (3) shows how the inclusion of the interaction terms affects the results. Respondents working in agriculture who report average quality of regional roads are estimated to have worked 1.7 hours less the previous week than people reporting good roads. Interestingly, people employed in agriculture who report poor regional roads 17 Column (5) reports the analysis at the household level. 18 are estimated to work slightly more than agricultural employees who report good regional roads. They are estimated to have worked 3.5 hours more the week before. These results suggests that while people employed in agriculture indeed work for fewer hours, they work more if they have poor connectivity with towns and markets. The reasons could be that people with inadequate access have difficulties accessing agricultural inputs and extension services, therefore are less productive and need to do more manual work. 19 Table 2: Marginal effects - probability of positive outcome. Non-agricultural Seasonal employment Job outside village employment (1) (2) (3) (4) (5) Poor reg. road quality -0.054*** -0.035*** -0.054*** -0.052*** -0.028* (0.006) (0.008) (0.007) (0.009) (0.015) Average reg. road quality -0.019*** -0.014** -0.015*** -0.015** -0.033*** (0.005) (0.006) (0.006) (0.007) (0.011) Log distance to market -0.013*** -0.023*** -0.015*** (0.003) (0.003) (0.005) Log distance to kindergarten -0.008*** -0.009*** -0.007** (0.002) (0.002) (0.003) Log distance health center -0.006** 0.012*** -0.006 (0.003) (0.003) (0.005) Individual controls Yes Yes Yes Yes Yes Household controls Yes Yes Yes Yes Yes Region-year FE Yes Yes Yes Yes Yes Survey month FE Yes Yes Yes Yes Yes Observations 49,443 37,245 49,442 37,244 7,228 Individual controls include: gender, age, age squared, household head, marital status, education categories. House- hold controls include: family size, number of children, family member abroad as an immigrant. Robust standard errors in parentheses clustered on household level. ***p<0.01, **p<0.05, *p<0.1 20 Table 3: OLS estimates on number of hours worked in previous week (1) (2) (3) Agriculture -16.51*** -16.06*** (0.220) (0.368) Average reg. road quality -0.995*** 0.444 (0.242) (0.379) Poor reg. road quality -0.653** -2.902*** (0.318) (0.561) Agriculture x Average reg. road quality -1.717*** (0.438) Agriculture x Poor reg. road quality 3.533*** (0.638) Individual controls Yes Yes Yes Household controls Yes Yes Yes Geographic controls Yes Yes Yes Region-year FE Yes Yes Yes Survey month FE Yes Yes Yes Const. 14.47 0.87 14.95 Observations 37,245 37,245 37,245 R-squared 0.372 0.213 0.375 Individual controls include: gender, age, age squared, hh head, marital status, education categories. HH controls include: family size, number of children, family member abroad as an immigrant. Geographic controls include: log distance to market, log distance to kindergarten, log distance to health center. Robust standard errors in parentheses clustered on household level. ***p<0.01, **p<0.05, *p<0.1 21 6.2 DHS This section reports the results of the analyses performed on the combination of DHS survey and the road survey. I mapped the GPS locations of DHS survey clusters together with the road survey vector dataset and other geographic raster data layers. Table 4 shows the estimates of the first-stage regression - following the equation (2). The dependent variable is the endogenous variable LogDistGRs - log of distance to the nearest road in good condition. It is instrumented with LogDistRoads1903s - log distance to the nearest primary military and postal road in 1903. Table 4: First-stage regression. Coeff. s.e Logdist. good road Logdist. Road 1903 0.423*** (0.0252) Logdist. regional center -0.316*** (0.0446) Logdist. capital city 0.612*** (0.0846) Altitude 0.00101*** (0.0001) Slope -0.0716*** (0.0199) Population Density 2015 0.00438*** (0.00025) Nightlights Composite -1.652*** (0.0971) Aridity -0.00689 (0.00665) Individual controls Yes Household controls Yes Region dummies Yes Observations 3,801 R-squared 0.350 First stage F-stat 186.38 Individual controls include: age, age squared, gender, years of ed- ucation, individual weights. Household controls include: owns land usable for agriculture, wealth index, number of children of age 5 and lower, number of family members. Robust standard errors in parentheses. ***p<0.01, **p<0.05, *p<0.1 Table 4 shows that LogDistRoads1903s has a strong, statistically significant correla- tion with the endogenous variable LogDistGRs . The First-stage F-statistics also prove that LogDistRoads1903s is a strong instrument. The coefficients in Table 5 represent the average marginal effects on the probability 22 of non-agricultural employment and skilled manual employment. The marginal effects are estimated on the whole sample as well as on the samples of only women and only men. Column (1) shows the average marginal effects of the probability to be employed in the non-agricultural sector. The relationship between the distance to the nearest road in good condition and non-agricultural employment, as expected, is negative: 1 unit increase in log distance to the nearest good quality road, or approximately 2.7 fold increase (e ≈ 2.71828) decreases the probability of being employed in non-agricultural sector by 5.7 percentage points. The effects are consistent for women and men, showing, respectively, -5.8 and-6.2 percentage points likelihood of being employed in the non-agricultural sector. Next, the study estimates the average marginal effects of the probability of being employed in skilled manual employment given distance to the nearest good quality road. The results are reported in columns (4), (5), and (6). Overall, one unit increase in log distance (or approximately 2.7 fold increase) decreases the probability of having skilled manual job by 5.1 percentage points. When the estimations are performed on men and women separately we see that average marginal effects for women are still negative but lose statistical significance, while the coefficient for men still holds at statistically significant at negative 6.6 percentage points. This could be explained by the distribution of men and women in skilled employment shown in Figure 7. Only 6% of surveyed working women are employed in skilled manual employment, compared to 31% of men. Table 6 shows the average marginal effects of the probability of being employed in seasonal employment, and the probability of receiving cash earnings. Overall, living fur- ther from a road in good condition is negatively associated with seasonal employment, 1 log increase in distance from a good road decreases the probability of working seasonally overall by 5.5 percentage points. These findings are in line with the results of the analysis done on the ILCS data, presented in Table 2. When marginal effects are calculated sepa- rately for women and men, the coefficient of the average marginal effect of women having seasonal jobs becomes positive but statistically not significant. However, the marginal effect of the probability of seasonal employment on men is statistically highly significant - one unit increase in log distance decreases the probability of having a seasonal job by 10.7 percentage points. Unfortunately, limitations of the employment data in DHS surveys do not allow me to study deeply the seasonality outcomes. The descriptive statistics show 23 that almost half (47%) of agricultural workers regard their employment as non-seasonal, working whole year round. The results could indicate that going further from a good quality road increases the probability that people regard their agricultural employment as non-seasonal. This could be because of lower probability of getting off-season employment. Lastly, the study estimates the probability of getting cash earnings given distance to a road in good condition. The results suggest that one unit increase in log distance to a good quality road decreases the probability of getting cash earnings by 5.6 percentage points. When average marginal effects are estimated separately for men and women, effect on men is low and statistically not significant. On the other hand, being close to good quality road seems very important for women - 1 log - or approximately 2.7 fold increase in distance to a good quality road - decreases the probability of getting cash earnings by 9.3 percentage points. Overall, shorter distance to good quality road seems to be decreasing agricultural employment, increasing skilled manual and seasonal employment, and seems to provide outside village job opportunities. Proximity to good quality roads is particularly important for women - living closer to a road in good condition increases their likelihood to be employed in non-agricultural sector and get cash earnings for their work. 24 Table 5: Marginal effects of distance to roads in good condition on non-agricultural and skilled manual employment Non-agricultural employment Skilled manual employment (1) (2) (3) (4) (5) (6) All Women Men All Women Men Logdist. good road -0.057** -0.058* -0.062** -0.051** -0.027 -0.066** (0.024) (0.033) (0.029) (0.021) (0.029) (0.029) Age 0.022** -0.003 0.036*** 0.006 -0.001 0.019 (0.010) (0.013) (0.013) (0.009) (0.008) (0.015) Gender (woman=1) -0.177*** -0.198*** (0.019) (0.018) Education in years 0.045*** 0.062*** 0.019*** -0.019*** -0.017*** -0.018*** (0.005) (0.007) (0.007) (0.004) (0.005) (0.007) Number of children under 5 0.029* -0.025 0.066*** 0.004 0.020 -0.015 (0.016) (0.023) (0.022) (0.014) (0.014) (0.023) Owns agricultural land -0.171*** -0.216*** -0.093* -0.016 0.008 -0.041 (0.043) (0.064) (0.052) (0.030) (0.042) (0.050) Other individual controls Yes Yes Yes Yes Yes Yes Geographical controls Yes Yes Yes Yes Yes Yes Region dummies Yes Yes Yes Yes Yes Yes Observations 1,885 981 904 1,885 968 904 Geographical controls include: Distance to Marz center, Distance to Yerevan, Altitude, Slope, Population density, Nightlights composite, Aridity. Additional individual and household controls include: Age squared, individual weights, HH wealth categories, family size. Robust standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01 Table 6: Marginal effects of distance to roads in good condition on seasonal employment and cash earnings Seasonal employment Cash earnings (1) (2) (3) (4) (5) (6) All Women Men All Women Men Logdist. good road -0.055** 0.010 -0.107*** -0.056** -0.093*** -0.016 (0.024) (0.038) (0.025) (0.024) (0.029) (0.033) Age -0.035*** -0.032** -0.034** 0.030*** -0.011 0.065*** (0.010) (0.015) (0.013) (0.010) (0.013) (0.014) Gender (woman=1) -0.063*** -0.245*** (0.020) (0.020) Education in years -0.034*** -0.038*** -0.024*** 0.066*** 0.073*** 0.035*** (0.005) (0.007) (0.006) (0.005) (0.008) (0.007) Number of children under 5 -0.032* -0.019 -0.033 0.017 -0.042* 0.046** (0.017) (0.025) (0.022) (0.017) (0.022) (0.024) Owns agricultural land 0.015 -0.090 0.039 -0.171*** -0.239*** -0.083 (0.040) (0.063) (0.050) (0.044) (0.059) (0.056) Other individual controls Yes Yes Yes Yes Yes Yes Geographical controls Yes Yes Yes Yes Yes Yes Region dummies Yes Yes Yes Yes Yes Yes Observations 1,869 966 890 1,884 981 903 Geographical controls include: Distance to Marz center, Distance to Yerevan, Altitude, Slope, Population density, Nightlights composite, Aridity. Additional individual and HH controls include: Age squared, individual weights, HH wealth categories, family size. Robust standard errors in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01 Figure 5: Plotted margins coefficients 6.3 Robustness checks Primary and secondary roads Road quality might be related to the type of road, whether it is a major international road or a local road. Testing the distance to a primary road poses several challenges. First, no complete dataset exists on primary and secondary roads on Armenia. The latest complete dataset that could be obtained was from the Global Roads Inventory Project (GRIP), gathered, harmonized and integrated from almost 60 different geospatial datasets by the team of Meijer et al. (2018). However, the data on the South Caucasus, as well as other post-Soviet states, still remain incomplete. This could explain the high correlation (0.46) between primary roads from the GRIP dataset and historical roads from 1903, meaning that mostly old major roads might be captured in the GRIP dataset. This view is supported also by the low correlation between primary roads from the GRIP dataset and good quality roads (-0.02). The second problem is that, since the dataset has only few roads, including them in the regression together with the instrumental variable LogDistRoads1903s will result in multicollinearity. Multicollinearity might cause two basic problems: first, coefficient estimates become more sensitive to small change and 27 might swing highly, and second, multicollinearity reduces the precision of the estimated coefficients, weakening statistical power. These problems are evident from the regression results once the incomplete roads dataset is integrated in the analysis: while coefficients keep the same sign, the coefficient value and statistical power fluctuate highly. Location of mines Location of historical roads might be influenced by other factors than geography and military strategies. For instance, roads could have been built in areas with greater eco- nomic potential. During the period, which the historical map is depicting, the main non-agricultural sector with economic potential in Armenia was mining. Armenia has a long history of metal mining and is accounting a large part of the economy. According to the UN Comtrade, metals and ore concentrates accounted for more than half of Armenia’s exports in 2017.18 Copper mining first started in the Alaverdi area of Lori Marz in 1770s. Later, it started in Kapan in Syunik Marz in 1840s and by the mid-20th century the Kajaran copper-molybdenum mine in Syunik Marz started production. Kajaran mine is the largest operating mine in Armenia, accounting for 60% of the total mining turnover in the country (World Bank, 2016). In spite of the historical importance of the mining sector in Armenia, it is unlikely that by the end of the 19th century primary road placement would have been influenced by mine locations. There are two main reasons to argue this. First, most of the historical mines have been operated in Lori (in north Armenia) and Syunik (in south Armenia) regions of the country. In both regions, as 6 shows, large mines are located very far from the historical routes. However, they are connected by railway lines. The only mine which seems to be on the primary historical road route is the Ararat mine. However, the operation of the Ararat mine had been relatively small until 1975. In addition, it is not an active mine but rather a processing plant.19 18 UN Comtrade Database https://comtrade.un.org/data/ (Reviewed: December 2020). 19 Source: State Committee of Real Estate Cadastre of Armenia. 28 Figure 6: Mine locations in Armenia Source: Author’s compilation based on data on road quality from the World Bank, admin- istrative data and mine location from the Acopian Center for the Environment, historical roads from Military-topographic map of the Caucasus, 1903 Archives of the Russian State Library. 7 Discussion and conclusion Investing in improvement and maintenance of roads is widely believed to be vital for economic development. As a result, there is increasing pressure to invest in road infras- tructure, especially in low- and middle-income countries. There is a growing literature, studying the link between road improvement, reduced transportation cost and different economic outcomes. Nevertheless, there is a gap in identifying strong evidence on the link- age between road quality and economic outcomes. Moreover, evidence has been lacking on how road infrastructure quality affects the urban perimeter and impacts the employ- ment of the rural population and affects structural transformation. This paper attempts to study these relationships. This analysis uses two different sets of data and two different estimation methods to illustrate the link between road quality and rural employment. The first method uses 10 29 years of repeated cross-sectional Integrated Living Conditions Survey of individuals from 2007 to 2016, where respondents answered a set of questions about local infrastructure quality. The second method uses the Demographic and Health Survey from 2015-2016. Since DHS allows for approximate household location identification, I match the locations with a unique road quality dataset and use topographic control variables. In order to address the problem of endogeneity and reverse causation, the study uses an instrumental variable strategy based on historical military and postal routes in Armenia from a 1903 map, when Armenia was part of the Russian Empire. The analysis with both datasets and methods shows that road quality is positively associated with non-agricultural employment in rural areas. Households living further from good quality roads are more likely to be employed in agriculture and less likely to be employed in seasonal employment. The DHS analysis shows negative association between distance to good quality roads and employment outcomes. People are 5.7 percentage points less likely to work in the non-agricultural sector and 5.1 percentage points less likely to be engaged in skilled manual employment with 1 log increase in distance (or approximately 2.7 fold increase in distance). Women are particularly affected in the type of pay they receive for work. The results show that 1 log increase in distance decreases the probability of getting cash earnings for work by 9.3 percentage points. These results are particularly interesting in the prism of women’s empowerment, where financial independence is one of the key factors. The result on seasonal employment is particularly interesting for its unexpected op- posite sign. The analysis from both datasets shows a positive association of seasonal employment with better quality roads. The analysis on ILCS shows that people reporting poor road quality leading to towns and markets are 5.2 percentage points less likely to engage in seasonal work. The analysis on DHS shows 5.5 percentage points less likelihood of seasonal employment with one unit increase in log distance from a good quality road (approximately 2.7 fold increase). These results could be explained by the outcome that people living nearby improved roads are more likely to work outside the village. Better quality road infrastructure stimulates mobility and commuting through two main chan- nels: increased productivity, and reduced job search and commuting costs. Therefore, people living closer to better roads might be more likely to find additional jobs during less 30 active agricultural seasons. The paper is in line with the literature on agricultural productivity in terms of hours worked. Similar to McCullough (2017), this paper finds that agricultural workers work on average fewer hours than non-agricultural workers, 16.5 hours less per week. The results are very policy relevant. People living further away from good quality roads tend to be underemployed (working fewer hours), are less likely to work outside of village, and less likely to find seasonal jobs during inactive agricultural seasons. People who report average road quality in their region tend to work fewer hours in agricultural jobs than people living closer to good quality roads; however, respondents reporting poor quality roads work more hours in agriculture. This result could be related to access to agricultural inputs and productivity. People with average quality roads still have relatively adequate access to agricultural inputs, while poor quality roads hinder access to inputs, requiring more manual work. While many low- and middle-income countries are investing heavily in new road con- struction projects, keeping local roads in good condition should also be their priority. The literature shows that having higher maintenance expenditure rather than new infrastruc- ture investment can lead to a positive impact on output (Rioja, 2003a,b). Access to better quality road infrastructure is positively associated with rural em- ployment. The results are very policy relevant in the sense of structural transformation, providing non-agricultural jobs and new or additional work opportunities to people living in rural areas. 31 References Adamopoulos, T. (2011), ‘Transportation Costs, Agricultural Productivity, and Cross- Country Income Differences’, International Economic Review 52(2), 489–521. Aggarwal, S. (2018), ‘Do rural roads create pathways out of poverty? Evidence from India’, Journal of Development Economics 133, 375–395. Andrabi, T. and Kuehlwein, M. (2010), ‘Railways and price convergence in British India’, The Journal of Economic History 70(02), 351–377. Asher, S. and Novosad, P. (2016), ‘Market Access and Structural Transformation: Evi- dence from Rural Roads in India’, Manuscript: Department of Economics, University of Oxford . Asher, S. and Novosad, P. (2020), ‘Rural Roads and Local Economic Development’, Amer- ican Economic Review 110(3), 797–823. Banerjee, A., Duflo, E. and Qian, N. (2012), On the road: Access to transportation infrastructure and economic growth in China, Technical report, National Bureau of Economic Research. Banerjee, A. V. and Newman, A. F. (1998), ‘Information, the Dual Economy, and Devel- opment’, The Review of Economic Studies 65(4), 631–653. Baum-Snow, N., Brandt, L., Henderson, J. V., Turner, M. A. and Zhang, Q. (2012), Roads, railroads and decentralization of Chinese cities, Technical report, Working paper. Bell, C. and van Dillen, S. (2014), ‘How Does India’s Rural Roads Program Affect the Grassroots? Findings from a Survey in Upland Orissa’, Land Economics 90(2), 372–394. Brueckner, J. K. (1987), Chapter 20 The structure of urban equilibria: A unified treatment of the muth-mills model, in ‘Handbook of Regional and Urban Economics’, Vol. 2, Elsevier, pp. 821–845. o i Miquel, G. (2015), ‘The Burgess, R., Jedwab, R., Miguel, E., Morjaria, A. and Padr´ Value of Democracy: Evidence from Road Building in Kenya’, American Economic Review 105(6), 1817–1851. Datta, S. (2012), ‘The impact of improved highways on Indian firms’, Journal of Devel- opment Economics 99(1), 46–57. Deichmann, U., Shilpi, F. and Vakis, R. (2009), ‘Urban Proximity, Agricultural Potential and Rural Non-farm Employment: Evidence from Bangladesh’, World Development 37(3), 645–660. 32 Dercon, S., Gilligan, D. O., Hoddinott, J. and Woldehanna, T. (2009), ‘The Impact of Agricultural Extension and Roads on Poverty and Consumption Growth in Fifteen Ethiopian Villages’, American Journal of Agricultural Economics 91(4), 1007–1021. Donaldson, D. (2018), ‘Railroads of the Raj: Estimating the Impact of Transportation Infrastructure’, American Economic Review 108(4-5), 899–934. Duranton, G., Morrow, P. M. and Turner, M. A. (2014), ‘Roads and Trade: Evidence from the US’, The Review of Economic Studies 81(2), 681–724. Duranton, G. and Turner, M. A. (2012), ‘Urban Growth and Transportation’, The Review of Economic Studies 79(4), 1407–1440. Faber, B. (2014), ‘Trade Integration, Market Size, and Industrialization: Evidence from China’s National Trunk Highway System’, The Review of Economic Studies 81(3), 1046– 1070. Fafchamps, M. and Shilpi, F. (2003), ‘The spatial division of labour in Nepal’, The Journal of Development Studies 39(6), 23–66. Fafchamps, M. and Shilpi, F. (2005), ‘Cities and Specialisation: Evidence from South Asia’, The Economic Journal 115(503), 477–504. Fan, S. and Chan-Kang, C. (2005), Road Development, Economic Growth, and Poverty Reduction in China, number 138 in ‘Research Report’, International Food Policy Re- search Institute, Washington, DC. Garcia-L´ ´ Holl, A. and Viladecans-Marsal, E. (2015), ‘Suburbanization and opez, M.-A., highways in Spain when the Romans and the Bourbons still shape its cities’, Journal of Urban Economics 85, 52–67. Gertler, P. J., Gonzalez-Navarro, M., Gracner, T. and Rothenberg, A. (2022), ‘Road Maintenance and Local Economic Development: Evidence from Indonesia’s Highways’. Ghani, E., Goswami, A. G. and Kerr, W. R. (2016), ‘Highway to Success: The Impact of the Golden Quadrilateral Project for the Location and Performance of Indian Manufac- turing’, The Economic Journal 126(591), 317–357. Gibson, J. and Olivia, S. (2010), ‘The Effect of Infrastructure Access and Quality on Non-Farm Enterprises in Rural Indonesia’, World Development 38(5), 717–726. Gibson, J. and Rozelle, S. (2003), ‘Poverty and Access to Roads in Papua New Guinea’, Economic Development and Cultural Change 52(1), 159–185. Gollin, D. (2023), ‘Agricultural productivity and structural transformation: Evidence and questions for African development’, Oxford Development Studies 51(4), 375–396. 33 Gollin, D., Lagakos, D. and Waugh, M. E. (2014), ‘Agricultural Productivity Differences across Countries’, American Economic Review 104(5), 165–170. Gollin, D. and Rogerson, R. (2014), ‘Productivity, transport costs and subsistence agri- culture’, Journal of Development Economics 107, 38–48. Herrera Dappe, M., Alam, M. M. and Andres, L. (2021), The Road to Opportunities in Rural India, Mobility and Transport Connectivity, World Bank. Hodler, R. and Raschky, P. A. (2014), ‘Regional Favoritism’, The Quarterly Journal of Economics 129(2), 995–1033. Holl, A. (2016), ‘Highways and productivity in manufacturing firms’, Journal of Urban Economics 93, 131–151. Jedwab, R. and Storeygard, A. (2017), ‘The Average and Heterogeneous Effects of Trans- portation Investments: Evidence from sub-Saharan Africa 1960-2010’. Kebede, H. A. (2024), ‘Gains from market integration: Welfare effects of new rural roads in Ethiopia’, Journal of Development Economics 168, 103252. Khandker, S. R., Bakht, Z. and Koolwal, G. B. (2009), ‘The Poverty Impact of Ru- ral Roads: Evidence from Bangladesh’, Economic Development and Cultural Change 57(4), 685–722. Khandker, S. R. and Koolwal, G. B. (2011), Estimating the Long-Term Impacts of Rural Roads: A Dynamic Panel Approach, SSRN Scholarly Paper ID 1952496, Social Science Research Network, Rochester, NY. Lewis, W. A. (1954), ‘Economic Development with Unlimited Supplies of Labour’, The Manchester School 22(2), 139–191. Lokshin, M. and Yemtsov, R. (2005), ‘Has Rural Infrastructure Rehabilitation in Georgia Helped the Poor?’, The World Bank Economic Review 19(2), 311–333. Mathematica (2015), ‘Evaluation of a Rural Road Rehabilitation Project in Armenia’, https://www.mathematica.org/publications/evaluation-of-a-rural-road-rehabilitation- project-in-armenia. Mayala, B., Fish, T. D., Eitelberg, D. and Dontamsetti, T. (2018), ‘The DHS Program Geospatial Covariate Datasets Manual’, USAID p. 50. McCullough, E. B. (2017), ‘Labor productivity and employment gaps in Sub-Saharan Africa., Labor productivity and employment gaps in Sub-Saharan Africa’, Food policy, Food Policy 67, 133, 133–152. 34 Meijer, J. R., Huijbregts, M. A. J., Schotten, K. C. G. J. and Schipper, A. M. (2018), ‘Global patterns of current and future road infrastructure’, Environmental Research Letters 13(6), 064006. oller, J. and Zierer, M. (2018), ‘Autobahns and jobs: A regional study using historical M¨ instrumental variables’, Journal of Urban Economics 103, 18–33. Mu, R. and van de Walle, D. (2011), ‘Rural Roads and Local Market Development in Vietnam’, Journal of Development Studies 47(5), 709–734. Munshi, K. and Rosenzweig, M. (2016), ‘Networks and Misallocation: Insurance, Migra- tion, and the Rural-Urban Wage Gap † ’, American Economic Review 106(01), 46–98. Nakamura, S., Bundervoet, T. and Nuru, M. (2020), ‘Rural Roads, Poverty, and Resilience: Evidence from Ethiopia’, The Journal of Development Studies 56(10), 1838–1855. Nguyen, K.-T., Do, Q.-A. and Tran, A. (2011), One Mandarin Benefits the Whole Clan: Hometown Infrastructure and Nepotism in an Autocracy, SSRN Scholarly Paper ID 1965666, Social Science Research Network, Rochester, NY. Qin, Y. and Zhang, X. (2016), ‘The Road to Specialization in Agricultural Production: Evidence from Rural China’, World Development 77, 1–16. Redding, S. J. and Turner, M. A. (2015), Chapter 20 - Transportation Costs and the Spatial Organization of Economic Activity, in J. V. H. a. W. C. S. Gilles Duranton, ed., ‘Handbook of Regional and Urban Economics’, Vol. 5 of Handbook of Regional and Urban Economics, Elsevier, pp. 1339–1398. Rioja, F. K. (2003a), ‘Filling potholes: Macroeconomic effects of maintenance versus new investments in public infrastructure’, Journal of Public Economics 87(9–10), 2281–2304. Rioja, F. K. (2003b), ‘The penalties of inefficient infrastructure’, Review of Development Economics 7(1), 127–137. Shamdasani, Y. (2021), ‘Rural road infrastructure & agricultural production: Evidence from India’, Journal of Development Economics 152, 102686. Sharma, A. (2016), ‘Urban Proximity and Spatial Pattern of Land Use and Development in Rural India’, The Journal of Development Studies 52(11), 1593–1611. Sheng, Y., Zhao, Y., Zhang, Q., Dong, W. and Huang, J. (2022), ‘Boosting rural labor off- farm employment through urban expansion in China’, World Development 151, 105727. Sotelo, S. (2015), ‘Domestic Trade Frictions and Agriculture’, Working paper p. 55. Stifel, D. and Minten, B. (2017), ‘Market Access, Well-being, and Nutrition: Evidence from Ethiopia’, World Development 90, 229–241. 35 Volpe Martincus, C., Carballo, J. and Cusolito, A. (2017), ‘Roads, exports and em- ployment: Evidence from a developing country’, Journal of Development Economics 125, 21–39. unen, J. H. (1826), Isolated State, Vol. An English edition of Der isolierte Staat, Von Th¨ Pergamon, English Translation, 1966 by Peter Geoffrey. Wang, X., Zhang, X., Xie, Z. and Huang, Y. (2016), Roads to Innovation: Firm-level Evidence from China, Vol. 1542, International Food Policy Research Institute (IFPRI). Wiegand, M., Koomen, E., Pradhan, M. and Edmonds, C. (2017), ‘The Impact of Road Development on Household Welfare in Rural Papua New Guinea’, p. 33. World Bank (2016), Armenia - Strategic mineral sector sustainability assessment, Techni- cal Report 106237, The World Bank. World Bank (2017), Connecting the dots : Transport, poverty, and social inclusion - evidence from Armenia, Technical Report 125193, The World Bank. World Bank, T. (2009), World Development Report 2009 : Reshaping Economic Geogra- phy, The World Bank. 36 8 Appendix Table 7: Summary statistics of variables used - ILCS Survey Variable Variable description Obs. Mean Sd. Min. Max. Outcome Dummy: respondent is employed in nagr 49,442 0.28 0.45 0 1 non-agricultural employment Dummy: respondent is employed in season 49,442 0.27 0.44 0 1 seasonal employment Dummy: household member works outjobdum 46,661 0.14 0.34 0 1 outside village Hours worked during previous hours 49,209 28.42 16.45 0 168 week Controls Road quality categories: 1 - rqreg 49,442 1.93 0.68 1 3 good, 2 - average, 3 - poor sex Gender of a respondent 49,442 0.51 0.50 0 1 Dummy: respondent is a household hhhead 49,442 0.30 0.46 0 1 head age Respondent’s age 49,442 44.00 14.08 16 85 agesq Age squared 49,442 2134 1272 256 7225 marstatus Marital status of a respondent 49,442 1.35 0.73 1 5 Highest education achieved by a edu 49,442 3.32 0.51 1 4 respondent members Number of family members 49,442 4.94 1.82 1 16 Dummy: household member is fammigrwork 49,442 0.17 0.38 0 1 working abroad Number of children in the family childcount 49,442 1.03 1.12 0 7 (under the age of 16) Number of elderly people age 65 eldercount 49,442 0.47 0.69 0 4 and higher in the family market Distance to the nearest market 37,250 16.28 12.32 0.01 100 Distance to the nearest kinder 37,269 4.46 7.04 0 52 kindergarten Distance to the nearest health healthc 37,272 1.10 5.07 0.002 500 center 37 Table 8: Summary statistics of variables used - DHS Survey Variable Variable description Obs. Mean Sd. Min. Max. Outcome Dummy: respondent works in nagr 1,885 0.64 0.48 0 1 the non-agricultural sector Dummy: respondent has skilled smanual 1,885 0.18 0.38 0 1 manual employment Dummy: respondent works season 1,869 0.34 0.47 0 1 seasonally Dummy: respondent has cash cash 1,884 0.57 0.50 0 1 earnings Treatment Distance to the nearest DistGR 1,885 2.03 2.11 0.00 13.28 good/very good quality road IV Distance to the nearest 1903 DistRoad1903 1,885 8.31 10.82 0.08 59.01 year primary road Controls age Respondent’s age 1,885 34.89 8.46 16 49 agesq Age squared 1,885 1289.07 590.92 256 2401 edu Respondent’s years of education 1,885 10.95 2.29 0 20 gender Dummy: Gender of a respondent 1,885 0.52 0.50 0 1 weights Individual survey weights 1,885 991506.2 333707.5 262419 1998569 wealth Wealth index categories 1,885 1.98 1.00 1 5 familysize Number of family members 1,885 4.90 1.66 1 11 Number of children in the family child5 1,885 0.40 0.70 0 4 of 5 years old and below Dummy: household holds an agland 1,885 0.92 0.27 0 1 agricultural land Distance to the regional (Marz) dist marz 1,885 18.30 12.19 0.99 63.63 center dist yerevan Distance to the capital city 1,885 58.66 43.12 6.85 209.15 ALT DEM Elevation of a settlement 1,885 1320.26 460.98 594 2331 slope Slope of a settlement 1,885 3.41 2.70 0.10 12.40 pop dens Population density 1,885 234.12 453.59 22.51 2521.60 Night-time light composite over nightlight 1,885 0.60 1.21 0 5.809831 the settlement Aridity of land (measured in aridity 1,885 18.56 4.65 11.23 29.38 2015) Note: DistGR, DistRoad1903, Dist marz , and dist yerevan are calculated in ArcGIS using NEAR function, and are used in log form in estimations. 38 Figure 7: Rural employment indicators by Gender (DHS) 39 Figure 8: Distribution of road network quality by region (Marz). Calculated based on the WB RoadLabPro data on road quality. 40 Figure 9: Military-topographic map of the Caucasus, 1903 Source: Archives of the Russian State Library. (Map reprinted in 1914.) 41