Infrastructure for Agriculture: Water, Electricity, Roads

1. Infrastructure for Agriculture: Water, Electricity, Roads Ahmed Mushfiq Mobarak Yale University

2. Evolution of this talk • Originally I was asked to present on “Strategies:Infrastructure, Inputs, AAS and Irrigation”. Including: • “Alternative strategies for investing in agriculture, promoting agricultural growth, investing in irrigation, investing in complementary infrastructure, subsidizing or increasing the availability inputs, the role of agricultural extension services, Green Revolution” • Seemed like a lot for 30 minutes (or even 330 minutes) • Luckily, negotiated to Investing In Infrastructure • What do we know about returns to investments in large infrastructure, like water, electricity, roads? • Short Answer: Not Much • (Now I wish I could talk about those other things!)

3. An important but difficult question • So, why don’t we know much about the returns to investment in say, electricity? • We can contrast areas with and without electricity • We could compare areas before versus after getting a connection to the grid • Unfortunately, electricity grid placement is decidedly based on demand projections • In health, education we can design small-scale randomized evaluations. More difficult in infrastructure • But, we need to be informed about how to allocate funds to buying textbooks, or hiring nurses versus building roads or extending the electricity grid

4. An Idea for a Non-Experimental Method Yellow dots = electricity generation plants

5. Source: USGS

6. The Power of Hydro in Brazil • Brazil has 13% of the world’s water • 89% of Brazil’s Electricity is provided by Hydropower. Geologic requirements: • Gradient • Water

7. “Quasi-experimental” Strategy • Let’s extract only the portion of the variation in electrification due to geologic factors (and ignore the portion due to socio-economic drivers) • Engineer given a budget every decade equal to the actual electricity infrastructure investment for all of Brazil that decade • Engineer asked to allocate that budget spatially by building hydro plants and transmission lines in a cost-minimizing manner. • The model’s allocation based solely on geologic considerations, without regards to any socio-economic considerations. • Examine the effects of this geologic variation (leading to differences in electricity access) on incomes, population movements etc.

8. Actual Electricity Evolves over time… 1970s 1960s

9. Actual Electricity Evolves over time… 1990s 1980s

10. As does the predicted electricity from the model 1960s 1970s

11. Findings and Mechanisms • Very large correlations between electrification and population density don’t survive in the “quasi-experimental approach” • Positive effect on local GDP per capita does survive • Why and how does electricity affect GDP? • Induces people and firms to move to electrified areas • Induces in-migration of highly productive people and firms • Changes firms’ production techniques • The mechanism matters a lot, because the policy implications are very different! • Let’s not just evaluate whether a project works, let’s figure out why, how and when it works

12. Another Example of the Importance of Mechanisms: Technology Adoption • Examples abound in development of relatively inexpensive welfare-improving technologies that are not adopted (Insecticide treated bed-nets, new seeds, fertilizer, hand-washing) • Why not? • Lack of liquidity, or a credit market failure • Lack of information about consequences • “Taste” and tradition • Disentangling different reasons for adoption has important policy implications – do we need to subsidize, or address risk aversion, or taste, or an information failure? • My project randomly assigns a variety of incentives to induce improved cook-stove adoption in rural Bangladesh

13. Generating the “theory”

14. Traditional stove Improved stove 1: Portable Improved stove 2: Chimney

15. Some of the Interventions • Demand very elastic for portable stoves (50% subsidy increases take-up by 62%). Price matters little for chimney stoves • Implies that households care a lot about keeping women healthy, but less willing to pay money to save their time. • Using two different stoves teaches us something more fundamental about household preferences, and this is key

16. Opinion Leader Influence • For portable stoves: • 21 percentage point increase in take-up if all opinion leaders chose to buy our stove, • 35 percentage point decrease in take-up if all opinion leaders rejected our stove. • Much smaller (almost zero) effects of opinion leader choices with chimney stoves • Why? Value added of chimney stove more easily observable directly • Adoption by rich people has a much stronger effect than adoption by the educated or political leader. • We learn how to use leaders and social networks to design a large-scale stove distribution project.

17. Message • Even with randomization, you can have a better project if you take advantage of pre-existing differences in institutions (type of leader) and technologies (type of stove or seed or fertilizer) • For example, we can learn a lot if we vary project sites where in some cases there is no water and in other cases there is easy access to water • If we observe differences across the two areas, that teaches us something fundamental about the underlying mechanism. We get better guidance when we replicate.

18. Another Example from the Water Sector

19. Wastewater Irrigation in Urban Ghana Consumers eating raw salad components

20. Let’s understand the mechanisms before we design the project • Can we simply train farmers to behave better? • What incentive to they have to undertake the effort? • Can we introduce a certification system? • Maybe in a supermarket, but for street side sellers? • Let’s give farmers something they want, and add some conditionalities to that subsidy. • What do farmers want?

21. Farmers often form informal groups and pump water up close their project Let’s create or subsidize this project, and add conditions to our subsidy. We’ll randomize the ownership structure and conditional transfers to understand when and how these groups work Understanding how these collectives function has broader implications for you and for your colleagues working in other sectors and countries.

22. Messages • Be creative and open-minded in project design. • Go beyond: “let’s give a subsidy, and see what happens.” Or “let’s provide extension services and see if it works” • There might be interesting variation in who communicates with whom (due to income or ethnic differences) • We can compare different ways of providing extension services and information

23. What About Macro Policies? • Infrastructure Provision and Management are macro-level decisions. • We cannot randomize macroeconomic policies being made at the national level, or state level or district level • For example, if we decentralize management of water to the district, how do we measure effects on quantities available and district water quality?

24. Evaluating a Decentralization Policy • World Bank, U.N., F.A.O. promote decentralization in various forms to improve public service delivery (e.g. WDR 2004) • However, decentralizing decision-making may create spillovers and efficiency losses due to negative externalities: • The flow of rivers creates upstream and downstream jurisdictions • Upstream has an incentive to over-use and to pollute as the river is about to exit the jurisdiction. • In the absence of coordination in management, there may be over-pollution and over-use of water

25. Effects of Borders on Water Quality Amazon Should not rely on cross-sectional differences to identify effects Sao Paulo

26. 1991 map D F A X B We want to hold location fixed and examine changes over time in the same river segment. Ordinarily would not have time series variation, except in Brazil these county boundaries change over time! (most often due to splitting) Number of counties increased from 4492 in 1991 to 5562 in 2001

27. 1994 map D F C A X B • Due to a split in 1994, for the same pair of stations (B&F) • The number of border crossings changes from 1 to 2 • Upstream station B’s distance to the nearest exiting boundary decreases (from BD to BC)

28. BOD Direction of Water Flow -15 -10 -5 River County boundary Results -0.05 0.21 0.06 0.03

29. Message • Let’s be clever about taking advantage of existing institutional arrangements in designing the evaluation • Whether or not you randomize, it might be interesting to look at the “intersection” of a randomized program and pre-existing differences in geography, policies or institutions. Examples: • Borders • Rivers • Land Slope