Road Development and Population Mobility in Indonesia. March 2, 2016

Transcription

1 Road Development and Population Mobility in Indonesia Brian Thiede, Louisiana State University Clark Gray, University of North Carolina at Chapel Hill March 2, 2016 Paper prepared for the 2016 annual meeting of the Population Association of America ABSTRACT Geographic isolation from population centers and markets is often framed as an impediment to reducing poverty and spatial inequality. This claim has been used to justify considerable public expenditures on rural road development, yet the demographic implications of such projects have been given only limited attention in existing research. We address this gap by advancing two main objectives using data from four rounds of the Indonesian Family Life Survey. First, we estimate the effect of road development on the probability of within- and between-province migration among adults aged Second, we test whether the effects of road development on migration are modified by factors that commonly predict migration, including age, sex, educational attainment, household wealth, and remoteness. We also disaggregate migration by duration of move. Results show that construction of asphalt or concrete roads increase migration, but only within provinces. We also find evidence that road-induced migration is generally concentrated among individuals with low educational attainment and wealth, and those living in communities most distant from their provincial capital. Results also suggest that most road-induced moves are temporary in nature.

2 1. INTRODUCTION Geographic isolation from population centers and markets is often framed as a major impediment to reducing poverty and spatial inequality. This claim has been used to justify considerable public expenditures on rural road development (Bird et al. 2010, Riverson et al. 1992, World Bank 2009). The construction of roads like other spaceand time-compressing technologies is expected to improve the economic status of affected populations by decreasing transportation and transactions costs, and increasing market integration (Key et al. 2000, Barrett 2008). While a growing body of empirical research supports many of the theoretical claims regarding roads economic impacts (Jacoby and Minten 2009, Mu and van de Walle 2011, Stifel and Minten 2008, Stifel et al. 2012), less attention has been paid to the demographic implications of road development (see Beauchemin and Schoumaker 2005, Gachassin 2013, Lichter and Fuguitt 1980 for exceptions). The lack of attention to effects on migration is particularly surprising given the responsiveness of migration to changing economic conditions and spatial-temporal barriers on the one hand; and the consequences of selective in- and outmigration for affected communities on the other. We address this gap by advancing two main objectives using data from the Indonesian Family Life Survey (IFLS). First, we estimate the effect of road development on the respective probabilities of within- and between-province migration among adults aged in Indonesia. Second, we test whether the effects of road development on migration vary according to factors upon which migration is commonly selected and according to which the social and economic effects of road development are likely to be distributed. Assessing potential selectivity in road-related migration patterns will provide insight into how the demographic effects of road development will affect the social and economic composition of communities targeted by road development. Since the marginal effects of road improvements likely vary by distance to the population centers that roads may link communities to, we also assess whether the effects of road development on migration vary according to the remoteness of affected persons residence from provincial capitals. Our analyses underscore the need to assess the demographic effects of road development alongside economic effects: demographic changes have important consequences in their own right and may have second-order effects on economic development. 2. ROAD DEVELOPMENT AND POPULATION MOBILITY Governments and inter-governmental organizations across the world allocate large amounts of resources toward building and improving road infrastructure. These investments are predicated on the assumption that the resulting increase in spatial connectivity between major population centers and the previously isolated areas affected by road development will promote economic development (Gannon and Liu 1997, van de Walle 2002). These effects are largely expected to operate via a reduction in transportation and other transactions costs associated with large temporal and spatial barriers between economic actors. The increase in spatial connectivity due to road developments is believed to reduce these costs, making economic exchange between 2

3 previously distant actors more likely (Minten et al. 2013, Stifel and Minten 2008). The economics literature suggests that declining transactions costs increase the effective market price that producers earn, and lowers the effective price for consumers (Minten and Kyle 1999, Renkow et al. 2004). These costs shape households economic ability to participate in markets (versus autarkic production) and the production decisions of market-oriented producers (Key et al. 2000). The implication is that changes in transactions costs associated with improved spatial connectivity should be strongly linked to aspects of household livelihood, including where income-generating activities take place. A specific example of the effects of road development is the integration of commodity markets in previously isolated areas with those in major population centers. For instance, isolated rural markets for agricultural commodities may become linked into larger regional or national circuits of exchange. This integration may have positive effects for previously isolated households, such as gains stemming from improved access to productivity-enhancing inputs and diminished seasonal fluctuations in food prices (Stifel and Minten 2008). However, integration may also expose those same communities to potentially destructive competition and exogenous macroeconomic shocks (e.g. exchange rate fluctuations, global commodity price spikes), with adverse consequences for wellbeing (Barrett 2008). This one example is consistent with a more general acknowledgement that the effects of connectivity and reduced transactions costs are not unambiguous, but rather determined by the enabling environment (e.g., market institutions) and the characteristics of the affected populations (Barrett 2008, Bird and Shepherd 2003, Jacoby 2000, Mu and van de Walle 2011, Pinstrup-Anderson and Shimokawa 2006, Porter 2007). The hypothesized economic changes associated with road development have clear implications for migration, which is commonly (but not exclusively) motivated by economic concerns such as maximizing wages and mitigating risks. However, like expectations about the overall effects of road development, existing theories of migration provide competing expectations about how road development may affect population mobility (as Gachassin 2013). On the one hand, improved transportation infrastructure decreases the effective distance between areas connected by roads. This should lower transportation time and costs including the social and psychological costs of migration and thereby increase the odds of migration (Gachassin 2013, Greenwood 1997). Improved roads may also bring about new flows of people, ideas, and goods, which may increase the likelihood of migration by altering aspirations and norms, increasing flows of information, and expanding social networks among previously isolated and marginalized populations (Bird et al. 2010). On the other hand, the positive economic returns to road development cited above may decrease out-migration from newly connected areas by reducing wage and amenity gaps previously shown to affect migration (de Haas 2010, Todaro 1969). Previous research suggests that improvement in spatial connectivity via roads is positively associated with many development indicators, including agricultural wages, food availability, school completion rates, health indicators, and the development of markets and financial institutions (Binswanger et al. 1993, Chamberlin and Jayne 2011, Mu and van de Walle 2011). While such developments have indeed been framed as reducing the incentives for migration, changes such as increased wages and education may eventually 3

4 promote migration by increasing employment prospects and providing resources to fund longer-distance moves. Given these conflicting but plausible expectations for whether and how road development will affect out-migration, an empirical approach is needed. Few studies have addresses this question to date, and existing evidence provides a range of results. For example, Beauchemin and Schoumaker (2005) examine how a broad set of development indicators including road development affect migration to cities in the case of Burkina Faso. Contrary to expectations that development in the place of origin will reduce out-migration, their results suggest either null effects or increased likelihood of mobility. With respect to road development in particular, this study finds that residence in a rural village connected to an all-weather road has a strong positive effect on the likelihood of out-migration to cities. Other studies find the opposite. For example, a recent study in the Tanzanian context found that road improvements reduced migration, showing evidence that this effect was due to improving conditions among those who remained in affected communities (Gachassin 2013). Consistent with these findings, Shilpi et al. (2014) find that migrants in Nepal are more likely to relocate in areas near improved roads and with access to electricity. This finding is consistent with the idea that residential preferences are correlated with development indicators. While not looking at road development per se but rather a broader set of infrastructure indicators Bilsborrow and colleagues (1987) found that access to electricity and off-farm employment reduced migration in Ecuador. However, they also found that access to local urban environments increased the likelihood of out-migration to more distant areas. This latter finding is consistent with the expectation that increased exposure to new social and cultural norms may promote future migration, particularly among previously isolated populations. Aside from this limited set of studies, we know of little other research that addresses the effects of improved road infrastructure on human migration. The dearth of research on this topic may in part reflect well-documented challenges of assessing the impacts of road development (van de Walle 2013). Yet knowledge of whether road development affects migration, if it increases or decreases out-migration from previously isolated areas, and how road effects on migration are distributed across the affected population, can all refine road impact assessments. Indeed, evidence of significant changes in migration associated with road development would suggest the possibility of large second order effects on infrastructure projects effects that have not been fully accounted for to date. 3. DATA AND METHODS We address our research questions using panel data from the Indonesian Family Life Survey (IFLS). The IFLS is a uniquely rich longitudinal study of over 30,000 Indonesians. The sample was originally selected to be representative of the population living in 13 of the 27 provinces in the country accounting for 83% of the country s population (Strauss 2009). With respect to the current analysis, the IFLS has the advantage of including sufficiently detailed information on both individual migration histories and community-level infrastructure. The migration histories were collected for 4

5 all household members aged 15 and above during the IFLS household survey. The community-level data were collected through interviews with the village head (lurah) and any staff that leader selected to assist with the interview. Records from the village office were also used to support interview data as needed. Regarding the focus of our study, the village leaders were asked to describe the prominent type of road in the village defined as the main street that linked their community with others. We also use information on other forms of infrastructure in the community (e.g., electricity, irrigation), and the distance between the community (measured from the village head s office) and provincial capital that were collected in this interview. Another key advantage of the ILS is its remarkably high rate of follow-up including among migrants thereby limiting attrition bias. For example, 84% of the living household members from IFLS 1 were interviewed in IFLS 2. For the following two inter-survey periods (IFLS 2 to IFLS 3 and IFLS 3 to IFLS 4), comparable rates of follow-up were 84.0% and 77.1%, respectively (Duncan et al. 2012). In addition to the quality of the IFLS, the Indonesian context also provides an appropriate case for studying this topic. Road conditions in the country have been characterized by significant variation across space and time. Spatially, consider that in 2010, less than half of the country s roads were in good condition, with majority of roads characterized as moderate to severely damaged (Parikesit 2006, quoted in Gibson and Olivia 2010). Recent conditions nonetheless represent a significant improvement from the past. The country s road network increased by nearly 20% between 2000 and 2010 (Gibson and Olivia 2010), and the Indonesian government continues to spend large amounts of money US$7 billion per year, or 40% of infrastructure spending on road infrastructure development (World Bank 2013). There is also evidence that these changes in infrastructure are affecting livelihoods: using a single round of IFLS data, Gibson and Olivia (2010) find that engagement in non-farm enterprises and income shares from such enterprises increases in step with improvements in road quality (and electricity). Since migration decisions occur within the broader context of livelihood strategies, evidence that roads are inducing changes in the both sector of employment and earnings justifies expectations of parallel migration effects. Using multiple rounds of the IFLS, we address this general question here. We do so by constructing a person-period dataset that includes indicators of between- and within-province moves, a set of individual and household characteristics known to affect migration (e.g., age, sex, household wealth), indicators of local road conditions, and measures of other dimensions of community infrastructure (e.g., presence of electricity, schools, irrigation, rural/urban status). To avoid endogeneity with the migration decision, predictors are measured at the baseline of each survey interval and migration is then measured over the survey interval. By controlling for other dimensions of development, we reduce the risk that estimated road effects are confounded by other simultaneous changes in infrastructure and development at the community level. We restrict our sample to adults aged at period baseline and who are observed in at least two consecutive panels of data. Our final analytic sample includes a total of 27,902 person-period observations. Following prior research on this topic (e.g., Beauchemin and Schoumaker 2005), we estimate a series of discrete-time event history models using logistic regression. Since we distinguish between within- and between-province moves, we estimate multinomial 5

6 models. We include community and period fixed effects in all specifications of the model. This approach addresses the endogenous road placement issue that has limited previous research on the impacts of road development, the locations of which tend to be targeted by planners (van de Walle 2009). Our statistical models takes take the general form: log!!"#!!"# =!!" +!!" +!!!!" where!!"# is the odds of migration outcome m for individual i in period t,!!"# is the odds of no migration during that period,!!" is the baseline liklihood of migration outcome m during period t,!!" is the baseline likelihood of migration outcome m in community c,!!" is a vector of individual, household, and community-level variables for individual i in the first year of period t, and!! is a vector of parameters for the effects of those independent variables on the odds of migration outcome m. We estimate a maineffects model as well as a series of interaction models that test for heterogeneous effects of road development on migration according to a set of individual-, household-, and community-level characteristics. 4. RESULTS 4.1. Descriptive statistics We begin with basic descriptive analyses of our data (Table 1), which summarize the demographic characteristics of our analytic sample and provide an overall benchmark of the incidence of migration. The overall migration rate was 11.7 per 100, with a large majority (83.1%) of those moves involving relocation within the province of origin. In contrast to the within-province migration rate of 9.7 per 100, the between-province migration was rare with a rate of only 2.0 per 100. One of our analyses also decomposes migration according to duration, defining return migrants as those who returned to their community of origin by the end of the inter-survey period in question. Our descriptive analyses reveal that most migrants return to their community of residence by the end of the inter-survey period: the rate of temporary migration was 8.9 per 100; while the rate of permanent migration (so defined) was 2.8 per 100. We highlight two additional descriptive findings. First, there is considerable variation in the remoteness of communities defined here as the distance (km) from the provincial capital. While the mean distance is km, observed distances range from 0.2 km to 900 km. 1 Second and finally, we highlight the distribution of observations according to the type of road that serves as the main conduit to and from their community. Approximately three-quarters of our observations had access to asphalt or 1 Documentation regarding missing values in IFLS 1 was ambiguous, so distances coded as 999km were treated as missing and replaced with the distances reported during IFLS2. Consistent with our interpretation of these values as missing, the corresponding distances reported in IFLS 2 were less than 700km in all cases. 6

7 cement roads. However, more than 23% of the observations had access to only dirt roads (7.9%) or those paved with stones, pebbles, or other loose materials (15.5%). As one indicator of change in this outcome over time the focus of our analysis consider that the share of the analytic sample with access to asphalt or concrete roads increased from 66.9% during period one to almost eighty percent (79.6%) in the final period (results not shown). This change was matched by declines in the population shares with access to dirt roads (13.0% to 5.3%) and roads paved with only stones or pebbles (18.6% to 13.9%). These results demonstrate that considerable changes in transportation infrastructure occurred during the time period considered in our analysis Road effects on within- and between-province migration Our first regression model estimates the effect of changes in road conditions on the respective likelihoods of within- and between-province moves (Table 2). Estimates show that transitions between dirt, loosely paved, and asphalt or cement roads have nonsignificant effects on the likelihood of migration between provinces. However, the construction of asphalt or cement roads does increase the likelihood of within-province movement vis-à-vis the dirt road reference category. The coefficient estimate of translates to a relative risk ratio (RRR) of That is, the shift from dirt to asphalt or cement roads increases the likelihood of within-province migration by approximately 34% a sizeable effect. That this road effect is observed among intra-province moves may suggest that road development promotes changes at the regional rather than national level. This is consistent with expectations that the largest changes in exchange (of goods, capital, and people) are likely to occur between communities directly connected by an improved road. The concentration of this effect on within-province moves is also consistent with the geographic reality of Indonesia: many inter-province moves involve crossing bodies of water, which road development will not have a direct effect on. Of course, it is also important to note that within-province moves are not necessarily of short distance: distances to between communities in our sample and provincial capitals are in some cases many hundreds of kilometers long and sometimes on another island Heterogeneity in road effects The average effect of road development on migration may mask heterogeneity according to individual, household, and community characteristics. Our discussion of previous research highlighted evidence that the distribution of economic returns to road development and the subsequent social and economic changes it causes is not necessarily equal. One s ability to capitalize on the structural changes induced by road development or, conversely one s vulnerability to such changes is likely to be contingent upon gender, education and other skills, and economic resources (e.g., access to capital). These factors are also commonly observed predictors of migration. In addition to these demographic and economic characteristics, we also consider variation in road effects according to the community s distance from the provincial capital, which we interpret as an indicator of remoteness. Since motorized transportation is presumably most important in distant areas where walking and non-motorized transport to major population centers is prohibitively difficult we would expect the demographic changes 7

8 caused by road development to vary according to remoteness. For this series of models, we present and discuss only the estimates for intra-province moves (Table 3). However, note that these interaction models include the same set of controls presented in Table 2, and are estimated as a multinomial logit. Complete results (not shown) show that road effects on the likelihood of inter-province migration are non-significant for all specifications. We begin by interacting roads and gender (Specification A), and find male-female differences that are not statistically different from zero. Among men, the effect of shifts to asphalt or cement are only marginally significant, but positive (β=0.351). The effect of this type of road is not statistically significant among women (results not shown). Likewise, estimated road effects are not statistically significant when allowing these effects to vary across the year age range considered in our analysis (Specification B). Individuals educational attainment may shape their exposure and ability to adapt to the structural changes brought about by road development. For example, education may be correlated with individuals primary occupation prior to road development, with the implication that those with certain levels of educational attainment will be disproportionately affected (or not) by road-related changes. Likewise, one s skills may determine their relative prospects in labor markets recently expanded by road development. Newly available opportunities in distant locations may be more attractive for some individuals than others according to their educational attainment. When allowing road effects to vary by educational attainment (Specification C), we find that statistically significant road-related increases in migration are concentrated among those with low (below 7 years) and high (12 years and above) educational attainment. Among the former, the coefficient estimate for shifts to asphalt or cement roads is This figure is equivalent to a relative risk ratio of 1.684, indicating a 68.4% increase in withinprovince migration among individuals with low educational attainment. Our estimates of road effects among individuals with 12 or more years of education are also positive (β=0.385, RRR=1.470, results not shown), but only significant at marginal levels. These results suggest that road-related transformations change the incentives for migration among low-skilled adults so as to increase the likelihood of within-province moves. Our results also suggest (with less certainty) that individuals with relatively high educational attainment are also more likely to move after road improvements. Despite similarly positive effects, these changes in migration patterns are unlikely to be driven by the same motivations among the low- and high-educational attainment groups. In the fourth specification (Specification D), we interact our road indicators with a measure of household wealth. We expect the level of household economic resources to moderate the effects of road development on livelihoods in general and migration in particular by determining the extent to which households can withstand disruptions associated with changing spatial connectivity and/or invest in new opportunities that structural transformations bring about. Estimates from this model indicate that road effects on within-province moves are concentrated among members of asset-poor households the effects of road development are non-significant for the second and third highest wealth terciles. Among the asset-poor those in the lowest wealth tercile construction of asphalt or cement roads increases migration by nearly 52% relative to the dirt road reference category (β=0.417, RRR=1.517). 8

9 The final interaction model we estimate allows for variation in road effects by remoteness, as indicted by distance to the provincial capital (Specification E). While remoteness may make individuals less likely to migrate in response to structural changes, the shift from dirt to improved roads may have the most dramatic effects (overall and on migration) in remote areas where motorized transport may be essential to reach large population centers. Estimates from this model show that the effects of road improvement on migration occur mainly among residents of communities that are remote from provincial capitals. Road improvements have a non-significant effect on migration among residents of the communities in the lowest two-thirds of remoteness to the capital cities. In contrast, the effect of such road improvements among the most-remote third of communities is both positive and statistically significant at conventional levels. The coefficient estimate of suggests that among residence of these relatively remote communities, transitions from dirt to asphalt or concrete roads increase the likelihood of within-province migration by more than 73.2% (RRR=1.732). This result suggests that mobility among the populations most constrained by lack of transport infrastructure is more responsive to improvements in this infrastructure Road effects on temporary and permanent migration As a final analysis, we disaggregate all moves (regardless of distance) by duration (Table 4). Specifically, we distinguish between movers who remained in the destination at the end of the inter-survey period from those who had returned to their community of origin. 2 Evidence that road development induces permanent relocation could suggest a displacing effect, whereby road-related changes in previously isolated communities undermine the viability of livelihoods in those communities. In contrast, shorter-term migration would suggest engagement with the wage labor market in population centers that road improvements link affected communities to. This dynamic would be consistent with roadinduced livelihood diversification rather than displacement. Our results suggest, albeit with considerable uncertainty, that road development increases temporary moves, so defined. The coefficient estimate (β=0.397) is only marginally significant, but indicates a considerable increase in temporary migration due to the construction of asphalt or concrete roads (RRR=1.487). The coefficient estimate for permanent moves is not significant at even marginal levels. 5. DISCUSSION AND CONCLUSION This study provides new evidence that road infrastructure development affects the incidence and pattern of out-migration from affected communities. Using four waves of data from the IFLS, we find that construction of asphalt or cement roads increases the probability of out-migration from affected communities. These infrastructure effects are limited to within-province moves. While this finding in part reflects the unique 2 This measure represents a rough proxy of duration since the actual time a migrant remains in the destination is constrained by when, in the inter-survey period, the move took place. For example, a migrant who moved just eight months prior to the end of the inter-survey period would be treated the same as a migrant who moved three years prior to the end of the inter-survey period and remained in the destination. 9

10 geography of Indonesia where road infrastructure does little to aid inter-province migration across bodies of water it may also have substantive implications. Specifically, road infrastructure may be more likely to increase mobility from previously isolated areas to relatively local population centers (e.g., a provincial capital) rather than more distant locales. This suggests a response intended to deal with new circumstances (advantageous or disadvantageous) without incurring the costs and risks of long-distance moves. Suggestive evidence that most road-related moves are temporary rather than permanent in nature is also consistent with this interpretation. Our analyses also highlight important sources of heterogeneity in road effects. Road-induced migration is concentrated among individuals with low socioeconomic status, as indicated by low educational attainment and membership in the lowest household wealth tercile. Although we are unable to determine whether these migratory responses tend to be in response to push or pull factors, this result does suggest that the changes associated with road improvements are felt most by the poor and disadvantaged. If one assumes that migration is associated with improved outcomes, then our findings here are consistent with prior work on the economic returns to roads, which has suggested a progressive distribution of benefits (Jacoby 2000). We also found that the effect of road development on migration is limited to residents of remote communities, as measured by distance from the provincial capital. The lack of effect among residents in the two-thirds of communities nearest to these capitals may indicate that the access to motorized transportation that roads facilitate is not a major barrier to migration in which case road improvements do little to change the status quo. In contrast, residents of more remote communities may be less likely to move to regional population centers in the absence of motorized transport (e.g., due to prohibitively long journeys by foot or non-motorized transport). Therefore, the migratory response is strong when such communities experience substantial improvements in road conditions. Robust evidence of a road effect on migration underlines the need to continue exploring the demographic effects on transportation and other infrastructure projects. The changes in livelihood strategies that migration entails may have significant consequences for migrants and both origin and destination communities. To the extent that such migration fuels future development (e.g., via remittances) (de Haas 2010), road-induced migration may spur secondary benefits above and beyond the initial economic gains typically measured in impact assessments. In addition to a continued focus on infrastructure-migration linkages, future research should endeavor to more fully understand how migration behaviors correspond to other changes in livelihood affected by road development. Such evidence may allow researchers to better understand the extent to which road-induced migration is being driven by push or pull factors. A second task for future research is to identify the types of destinations that these migration streams are directed toward. The characteristics of the destination may have significant implications for migrant outcomes and the impact of such migration streams on receiving communities. 10

11 REFERENCES Barrett, C Smallholder Market Participation: Concepts and Evidence from Eastern and Southern Africa. Food Policy 33: Beauchemin, C. and B. Schoumaker Migration to Cities in Burkina Faso: Does the Level of Development in Sending Areas Matter? World Development 33: Bilsborrow, R. T. McDevitt, S. Kossoudji, and R. Fuller The Impact of Origin Community Characteristics of Rural-Urban Out-Migration in a Developing Country. Demography 24: Bird, K. and A. Shepherd Livelihoods and Chronic Poverty in Semi-Arid Zimbabwe. World Development 31: Bird, K., A. McKay, and I. Shinyekwa Isolation and Poverty: The Relationship between Spatially Differentiated Access to Goods and Services and Poverty. ODI Working Paper No Binswanger, H., S. Khandker, and M. Rosenzweig How Infrastructure and Financial Institutions Affect Agricultural Output and Investment in India. Journal of Development Economics 41: Chamberlin, J. and T. Jayne Unpacking the Meaning of Market Access. Working Paper No Michigan State University, Department of Agricultural, Food, and Resource Economics. de Haas, H Migration and Development: A Theoretical Perspective. International Migration Review 44: Gachassin, M. C Should I Stay or Should I Go? The Role of Roads in Migration Decisions. Journal of African Economies 22: Gannon, C. and Z. Liu Poverty and Transport. Paper No. TWU-30. Washington, DC: World Bank. Gibson, J. and S. Olivia The Effect of Infrastructure Access and Quality on Non- Farm Enterprises in Rural Indonesia. World Development 38: Greenwood, M.J Internal Migration in Developed Countries. In M.R. Rosenzweig and O. Stark (Eds.), Handbook of Population and Family Economics. Amsterdam: North-Holland. Jacoby, H.G. and B. Minten On Measuring the Benefits of Lower Transportation Costs. Journal of Development Economics 89: Key, N., E. Sadoulet, and A. de Janvry Transactions Costs and Agricultural Household Supply Response. American Journal of Agricultural Economics 82: Lichter, D. and G. Fuguitt Demographic Response to Transportation Innovation: The Case of the Interstate Highway. Social Forces 59: Minten, B., B. Koru and D. Stifel The Last Mile(s) in Modern Input Distribution: Pricing, Profitability, and Adoption. Agricultural Economics 44: Minten, B. and S. Kyle The Effect of Distance and Road Quality on Food Collection, Marketing Margins, and Traders Wages: Evidence from the Former Ziare. Journal of Development Economics 60: Mu, R. and D. Van de Walle Rural Roads and Local Market Development in Vietnam. Journal of Development Studies 47:

12 Parikesit, D Infrastructure Support for Improving Rural Investment Climate. Paper for the Indonesia Rural Investment Climate Assessment. Pinstrup-Anderson, P. and S. Shimokawa Rural Infrastructure and Agricultural Development. Paper prepared for Presentation at the Annual Bank Conference on Development Economics, Tokyo, Japan. Porter, G Transport Planning in sub-saharan Africa. Progress in Development Studies 7: Renkow, M., D.G. Hallstrom, and D.D. Karanja Rural Infrastructure, Transactions Costs, and Market Participation in Kenya. Journal of Development Economics 73: Riverson, J., J. Gaviria, and S. Thriscutt Rural Roads in sub-saharan Africa: Lessons from World Bank Experience. World Bank Technical Paper No Washington DC: World Bank. Shilpi, F., P. Sangraula, and Y. Li Voting with their Feet? Access to Infrastructure and Migration in Nepal. World Bank Policy Research Working Paper No Washington, DC: The World Bank. Stark, O. and J.E. Taylor Relative Deprivation and International Migration. Demography 26: Stifel, D. and B. Minten Isolation and Agricultural Productivity. Agricultural Economics 39: Stifel, D., B. Minten, and B. Koro Economic Benefits and Returns to Rural Feeder Roads: Evidence from a Quasi-Experimental Setting in Ethiopia. International Food Policy Research Institute (IFPRI) Ethiopia Strategy Support Program II (ESSP II) Working Paper No. 40. Addis Ababa, Ethiopia: IFPRI. Strauss, J., F. Witoelar, B. Sikoki and A.M. Wattie. "The Fourth Wave of the Indonesian Family Life Survey (IFLS4): Overview and Field Report". April WR- 675/1-NIA/NICHD. Thomas, D., F. Witoelar, E. Frankenberg, B. Sikoki, J. Strauss, C. Sumantri, and W. Suriastini. (2012). Cutting the Costs of Attrition: Results from the Indonesia Family Life Survey. Journal of Development Economics 98: Todaro, M.L A Model of Labor Migration and Urban Unemployment in Less Developed Countries. The American Economic Review 59: World Bank In Ethiopia All Roads Lead to Development. Accessed 31 July 2013 at: van de Walle, D Choosing Rural Road Investments to Help Reduce Poverty. World Development 30: van de Walle, D Impact Evaluation of Rural Road Projects. Journal of Development Effectiveness 1:

13 TABLES Table 1. Summary of variables Mean SD Min Max Migration by distance No move Within-province Between-province Migration by duration No move Temporary Permanent Roads Asphalt or cement Paved (with stones, pebbles, etc.) Dirt Rivers or lakes Other Sex=female Age Education 6 or less years years years Household assets (1,000 rupiah) a 26, , ,243,000.0 Household owns non-farm business Farm household Number of schools in community Hours of electricity per day, most reliable source in communi Technical or semi-technical irrigation present in community Cottage industry present in community Factory or plant present in community Rural residence Distance to provincial capital (km) Period IFLS IFLS IFLS N 27,902 a Regression analyses use the natural log of household assets (rupiah) 13

14 Table 2. Coefficient estimates of multinomial logistic regression models predicting migration outcome by distance Within-province Between-province Roads a Asphalt or cement ** Paved b (with stones, pebbles, etc.) Dirt Sex=female *** Age *** *** Education 6 or less years 7-11 years *** ** 12+ years *** *** Household assets (ln, rupiah) Household owns non-farm business *** Farm household *** *** Number of schools in village Hours of electricity per day, most reliable source in village Technical or semi-technical irrigation present in village Cottage industry present in village Factory or plant present in village Rural residence Period IFLS 1-2 IFLS IFLS *** *** Constant Joint test, road coefficients χ 2 =13.57 *** χ 2 = *** Pseudo R Log pseudolikelihood N ,902 +p<0.10, **p<0.05, ***p<0.01 All models include community-level fixed effects; standard errors are clustered at the community level. a This variable also accounts for the effects of living in a community where (a) river and/or lakes or (b) an unspecified other pathway serves as the main route to/from other communities. For the sake of brevity we exclude these results from this and subsequent tables. b Refers to paving with stones, pebbles, etc. 14

15 Roads a Table 3. Coefficient estimates from multinomial logistic regression models with interaction terms predicting migration outcome by distance, within-province migration only a Specification A Specification B Specification C Specifiction D Specifiction E Asphalt or cement *** ** *** Paved b Dirt Sex (female) X roads Sex (female) X asphalt or cement Sex (female) X paved Sex (female) X dirt Age X roads Age X asphalt or cement Age X paved Age X dirt Education (=7-11 years) X roads Education (=7-11 years) X asphalt or cement ** Education (=7-11 years) X paved Education (=7-11 years) X dirt Education (=12+ years) X roads Education (=12+ years) X asphalt or cement Education (=12+ years) X paved Education (=12+ years) X dirt Wealth (=middle tercile) X roads Wealth (=middle tercile) X asphalt or cement Wealth (=middle tercile) X paved Wealth (=middle tercile) X dirt Wealth tercile (=highest) X roads Wealth tercile (=highest) X asphalt or cement Wealth tercile (=highest) X paved Wealth tercile (=highest) X dirt Distance to province capital (=lowest tercile) X roads Distance to province capital (=lowest tercile) X asphalt or cement ** Distance to province capital (=lowest tercile) X paved Distance to province capital (=lowest tercile) X dirt Distance to province capital (=middle tercile) X roads Distance to province capital (=middle tercile) X asphalt or cement Distance to province capital (=middle tercile) X paved Distance to province capital (=middle tercile) X dirt Joint test, road coefficients χ 2 = χ 2 = χ 2 =34.49 *** χ 2 =45.79 *** χ 2 =58.61 *** Joint test, interaction terms) χ 2 =1.14 χ 2 =11.93 ** χ 2 =29.36 *** χ 2 = *** χ 2 = *** Pseudo R 2 Log pseudolikelihood N , , , , ,902 +p<0.10, **p<0.05, ***p<0.01 All models include control variables presented in prior tables and community-level fixed effects; standard errors are clustered at the community level. a We estimated all interaction models with alternative reference groups (where applicable) but report only a single of coefficient estimates. b Refers to paving with stones, pebbles, etc. 15

16 Table 4. Estimates of multinomial logistic regression models predicting migration Permanent move Temporary move Asphalt or cement Paved (with stones, pebbles, etc.) Dirt Joint test, road coefficients χ 2 = χ 2 =14.05 *** Pseudo R 2 Log pseudolikelihood N ,902 +p<0.10, **p<0.05, ***p<0.01 All models include control variables presented in prior tables and community-level fixed effects; standard errors are clustered at the community level. 16