Economic Benefits of Rural Feeder Roads: Evidence from Ethiopia

Abstract

We estimate households’ willingness-to-pay for rural feeder roads in Ethiopia. Using purposefully collected data, we compare the economic behaviour of households by remoteness to estimate the benefits of access to feeder roads. Although we cannot definitively assert a causal relationship, we cautiously estimate that gravel roads have internal rates of return of 12–35 per cent. These results suggest that rural feeder roads may have relatively high rates of return even in unfavourable settings where (a) small-scale farmers have low levels of marketed agricultural surplus, (b) non-farm earning opportunities are negligible, and (c) motorised transport services are not guaranteed.

Acknowledgements

The authors express their gratitude to DFID, USAID, and DFATD for funding this research under the Ethiopian Strategy Support Program II (ESSP II), a collaborative programme of the International Food Policy Research Institute (IFPRI) and the Ethiopian Development Research Institute (EDRI). They also want to thank participants in seminars held at the Ethiopian Development Research Institute (EDRI), the Ethiopian Economics Association (EEA), the Centre for the Study of African Economies at Oxford University, and at Cornell University for valuable comments on an earlier draft. The data and Stata code used in this analysis will be made available to researchers upon request to the authors.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. Jedwab, Kerby, and Moradi (2014) also find that colonial era railroad construction led to the emergence of major urban areas that persisted even after the departure of colonial settlers and the decline of the railroads.

2. Various approaches to handling the issue of endogenous road/railroad placement include panel data methods (Dercon et al., 2009; Jedwab & Moradi, 2014; Khandker et al., 2009), difference-in-differences (Mu and van de Walle, 2011), propensity score matching (Lokshin & Yemtsov, 2005), regression discontinuity (Casaburi et al., 2013), and instrumental variables methods (Binswanger et al., 1993). Given the long-term nature of the effects of roads, the time frame for longitudinal data needs to be sufficiently long in addition to being initiated prior to the construction/rehabilitation of the roads (Mu and van de Walle, 2011; Jacoby & Minten, 2009).

3. Wondemu and Weiss (2012) also use Jacoby and Minten’s (2009) approach to estimate the impact of rural roads in Ethiopia. However, they use the approach as a framework which provides the basis for their reduced-form specification of farm households’ demand for rural infrastructure using the Ethiopian Rural Household Surveys. They do not provide a complete measure of benefits.

4. While this approach is consistent with Chandra and Thompson (2000), Michaels (2008), and Jacoby and Minten (2009), it does come at the cost of diminished external validity.

5. Some 50 per cent of households in the least remote quintile have cell phones, while no more than 1 per cent of households in the remainder of the sample have them.

6. Half way through the survey area from the town of Garasghe, it is possible to reach within two hours the town of Finjit which is also connected to Gonder by a mostly passable gravel road (depending on mudslides). However, households in the survey still use Atsedemariam as their major output market since the market in Finjit is small. Some communities do gain access to modern inputs from Atsedemariam through Finjit.

7. Due to an enthusiastic enumerator, one extra household was interviewed beyond the target of 850.

8. In follow up interviews, farmers also stressed the physical costs associated with making tiring trips.

9. To minimise measurement error in estimating travel times and costs, each household’s transport costs is calculated as the average cost of the household’s reported cost and the costs reported by its five nearest neighbours. The nearest neighbours are determined using the GPS coordinates for each household.

10. These countries are Kenya, Uganda, Tanzania, Zambia, Malawi, Mozambique, and South Africa (Dorosh et al., 2012b).

11. The exchange rate was roughly 17 Birr per US Dollar.

12. Note that if land and soil characteristics (and variation in these characteristics) are homogeneous across the study area, then outcomes such as household consumption, child health and food security are a consequence of transport cost-induced choices and behaviours. As such, comparing choices and outcomes across the transport cost gradient is not an appropriate test for homogeneity.

13. For each plot for each season (Belg and Meher), farmers were asked questions related to rain (more/less than normal and earlier/later than normal), frosts or hailstones (more/less than normal) and pests or disease (more/less than normal).

14. We consider some variables such as plot characteristics to be fixed and not subject to choices made by the farming household. As such, these variables are not netted out (that is they are added back into the residual) of the adjusted yield. Note further that altitude is also a fixed characteristic and was included in the initial estimations. However, because it was not statistically significant for any of the cereals, it was dropped from the models. Multiple interactions among the various shocks and inputs were also included in the models, but did not affect the predictions of residual productivity.

15. Note that the yields in the survey area are roughly similar to those found in other studies of the Ethiopian highlands (see for example, Yu, Nin-Pratt, Funes, & Gemessa, 2011).

16. It is worth noting that the substitution of labour for modern inputs in more remote areas does indeed leave household’s worse off. But this has no bearing on the inherent productivity of the land, which is an indicator of the exogeneity of remoteness. The trade-off between labour and modern input use is a consequence of the transaction costs associated with remoteness, not a determinant of them.

17. In previous work on Madagascar, the well-known inverse relationship between plot size and land productivity and larger plot sizes in more remote areas helped to explain lower agricultural productivity in remote areas (Stifel & Minten, 2008). In the present analysis, however, it is not too surprising that yields are similar across the remoteness quintiles (controlling for labour inputs and modern inputs) because the distributions of plot sizes in the sample do not differ across the remoteness quintiles.

18. The fact that village sizes in the survey area do not differ systematically over the transport-cost gradient (see Table 1) is consistent with this. Note also that higher education levels observed for less remote households are not itself an indicator of spatial sorting based on ability since these households systematically have greater access to schools that are located in the market town. In a population that is randomly distributed on ability over the sample space, those that are closer to schools would be expected to acquire more schooling, especially if the returns are expected to be higher as well.

19. This section is based largely on Jacoby and Minten (2009).

20. This follows from our assumption that household and hired labour are perfect substitutes.

21. Technically, also depends on the household’s labour and land endowments and should be expressed as . However, we maintain the simplified notation for ease of reading.

22. We can do this by relying on the composite commodity theorem because market prices are determined outside of the study area and any changes in the effective prices for each of the components are due to changes in transport costs alone. Hence we can think of τ as the common price for all of the commodities.

23. Of course, these changes in income could affect the optimal levels of consumption of own production (x) and of imported bulk goods (c) that appear in the demand for transport tonnage, DT(τ). We follow Jacoby and Minten (2009) here and assume zero income effects in consumption. The implications of this are discussed in note 24.

24. Note that assuming that the income effects on consumption are zero means that the demand for transport tonnage curve is effectively a Marshallian demand curve. In other words, we use an uncompensated demand curve to calculate consumer surplus instead of a compensated (Hicksian) demand curve. Jacoby and Minten (2009) point out that if this assumption holds, then the Marshallian consumer surplus , equivalent variation (EV) and compensating variation (CV) are identical. If the zero income effect assumption does not hold, however, then CV ≤ ≤ EV.

25. Benefit estimates roughly double if we use transport costs that include the opportunity cost of time.

26. The choice of cropping patterns may indeed change with reductions in transport costs (von Thünen, 1966). Omamo (1998), for instance, shows that cropping patterns in which households devote large shares of land and other resources to low-yielding food crops rather than high-yielding cash crops may indeed be an efficient response to high transport costs to product markets.

27. In addition to making trips to the market town to acquire the inputs, roughly two thirds of the sample households made more than one trip to the market town in order to arrange the purchase of the inputs. In ignoring these additional trips, we acknowledge that we underestimate the demand for transportation.

28. There appears to currently be ample supply of transport services. Based on the marketed agricultural surplus estimated for the study area, an average of four trucks per week transport agricultural products from Atsedemariam to markets beyond. This is also consistent with informal interviews conducted with traders in Atsedemariam prior to the survey, and with truckers after the survey.

29. Note further that a 50 per cent reduction in the 2 Birr per kilometre per quintal transport cost for the most remote households in our sample results in a cost that is considerably more than the 0.14 Birr per kilometre per quintal transport costs observed on average for similar sized trucks on trunk roads between the wholesale markets (based on a survey of the 31 major wholesale markets in Ethiopia conducted by the authors). While we do not expect the transport costs on a rural feeder road to drop all the way to the trunk road costs, we equally do not expect them to be seven-fold larger than these costs.

30. In certain instances there are incentives that push households with weak non-labour asset endowments and who live in risky agricultural zones into allocating household labour to non-farm activities. Although households frequently do turn to the non-farm sector as an ex ante risk reduction strategy, distress diversification into low-return non-farm activities is also observed as an ex post reaction to low farm income (Haggblade, 2007; von Braun, 1989). This differs from such factors as earnings premia from high productivity/high income activities may attract, or pull, some household labour into non-farm employment (Barrett, Reardon, & Webb, 2001; Dercon & Krishnan, 1996; Haggblade, 2007; Lanjouw & Feder, 2001; Reardon, Berdegué, & Escobar, 2001). These high-return non-farm jobs may serve as a genuine source of upward mobility (Lanjouw, 2001).

31. We also estimated models that included altitude as an explanatory variable in the second model. As with the yield models, altitude was dropped from the freight weight models because its estimates were not statistically different from zero.

32. One community member lamented that when complications arise, ‘we send pregnant mothers to the hospital, and they come back corpses’.

33. This is consistent with previous costs in Ethiopia. For example, in 2008/9, it cost 260,000 Birr on average to construct a kilometre of rural gravel roads (Ethiopian Roads Authority, 2009). This figure comes from the second year of the third phase of the Ethiopian Rural Sector Development Programme (RSDP), in which 445.7 million Birr was distributed for the construction of rural roads and bridges, from which 1710 km of rural roads were built. Given inflation, this is conservatively equivalent to 730,000 Birr per kilometre at the end of 2011.

34. In our IRR calculations, we do not account for negative externalities such as road accidents and environmental effects, nor positive benefits such as improved access to health facilities and schools.

35. Because households in the sample are not evenly distributed across the transport cost/distance gradient (see Figure 1), the average benefit is calculated as a weighted average where the weights are the probability of the household being in the particular transport cost bracket.

36. For example, the estimated IRRs are between 37.5 and 47.8 per cent for the Nazareth-Assela road, 12.2 and 18.5 per cent for the Woreta-Gob road, 35.2 and 48.5 per cent for the Adigrat-Adwa road, 22.3 and 31.3 per cent for the Nekemte-Mekenajo road, and 13.3 and 20.9 per cent for the Dera-Magna road (World Bank, 2003).

37. Recall that the approach to estimating benefits in this paper cannot be applied to trunk roads because of the assumption of a small market and exogenously determined prices. Rehabilitation of major trunk roads that reduce transaction costs has been found to affect equilibrium prices (Minten, Stifel, & Tamru, 2014).

38. Jacoby and Minten (2009) also find that the bulk of the benefits estimated in their Madagascar sample derive from non-farm earnings.