My fellow bloggers, first David and then Markus and Jishnu, started a discussion last week exploring the validity of success claims from development donors about their financed projects. I’d like to continue this discussion with a focus on one particular dimension – not on the veracity of the claim of project success per se, but whether any project success is actually due to the actions and inputs (i.e. money) of the donor.
A credible evaluation of project impacts gives us confidence that the project actions resulted in welfare gains. However there is yet one more necessary leap in logic to arrive at the statement that donor project financing brought about the estimated gains. Stated simply: did the committed funds actually finance the evaluated project? While this question might sound strange at first, if donor funds are perfect substitutes for government funds – after all a “common sense” perspective tells us money is easily reallocated across sub-accounts – then there is no guarantee that donors are truly funding the project they laud. In reality, if the government already intended to implement the project, the donors are actually funding the marginal project, i.e. the one that would have otherwise remained on the shelf if not for the additional donor resources.
So this type of causal attribution not only involves project success but, further, if donor funds “stick” to the project like flies to flypaper or rather slosh around in one common budget bucket. This is a longstanding concern of course (here is a 1998 paper from Shanta Devarajan and Vinaya Swaroop), but one with comparatively little micro-empirical evidence to help us sort through the issues.
Fungibility has an intuitive feel – after all if I found $100 on the street and then did my regularly planned grocery shopping, it would be odd to claim that the additional $100 led to my grocery purchases. (Although of course the looser budget constraint as a result of the found cash may have led to the purchase of extra chocolate.) However governments typically aren’t a super-sized equivalent of the unitary household. Instead they are composed of competing interests arrayed across sectors and regions with contrasting interests and endowments. Because of this, donor funds may very well stick to the intended use without subsequent reallocation. The “common sense” perspective mentioned above may be wrong. There is a theoretical basis for this – as modeled by John Roemer and Joaquin Silvestre, collective decision making processes with heterogenous agents can create flypaper effects.
Dominique van de Walle and Ren Mu look at the experience of a rural roads project in Vietnam and whether the local government reallocates funds away from communities that have been awarded the project towards those communities that weren’t so lucky. To account for the possibility of endogenous program placement they adopt a Propensity Score Matched Difference in Difference framework and find that indeed the communities in the project built and rehabbed significantly more road than communities outside the project. As there were no effects on other types of infrastructural investments, this is indeed evidence for at least a partial “flypaper effect” at the local level. Why partial? Well the project emphasized road rehabilitation and not new road construction; however project communities spent more money on new roads and not rehabbing older ones. Their conclusion: within the transport sector there may be fungibility, but apparently no such fungibility across sectors.
Adam Wagstaff also looks at the case of donor aid to Vietnam, this time in the health sector with an investigation of the fungibility of donor health aid targeted to selected provinces. Similar to the study above, Adam finds no evidence of intersectoral fungibility, again suggesting that funds stick to the sector to which they are targeted. But the results do suggest a high degree of intrasectoral fungibility across provinces – apparently the Vietnamese health ministry explicitly stated at the start of the project that they would reallocate provincial funding to compensate regions not targeted by the program. This compensating behavior, if unknown to the researcher, is a real concern for any impact evaluation where the counter factual is comprised by non-targeted regions. The standard counterfactual comparison even in a randomized study would be invalidated by such a behavioral response.
While fungibility in the real world can invalidate IE designs, as in Adam’s case above, smart IE design can also help identify conditions where fungibility is likely to occur. More work of the sort will help us understand not only whether a program or project is ultimately successful, but to what degree success can be trumpeted by the financiers. The limited evidence reviewed here suggests that donor funds to a sector stay in that sector although the activities financed in that sector may not be the ones intended by the donors.
A postscript: There is a parallel issue embedded in this discussion that concerns the role of development agencies. Are development agencies more valued for the funds they provide or the knowledge they impart? Resources and knowledge are often bundled together in the form of a development project. But if indeed funds are highly fungible it may be more efficient to unbundle them (as in a Development Policy Lending type vehicle, if you happen to be familiar with the World Bank forms of lending and granting). Proponents of the traditional project based lending and granting may counter that the degree of knowledge exchange is maximized through the intensified interactions fostered by project specific design and supervision activities. If the gains (including any spillovers) from such interactions are great enough then a project approach to lending may be more successful even if funds are largely fungible. It’s an interesting and important discussion.