7.4. Sensitivity due to the data base
Several authors have criticized the sensitivity of Burnside
and Dollar's results to small changes in the database. For instance, Easterly
et al. (2003), use exactly the same specification as Burnside and Dollar (1997)
but added new available data to the sample. Hence, the former database has been
updated to 1997 and earlier gaps have been filled. As a result, the
«aid×policy» interaction term becomes negative and
insignificant. In other words,
33 Tropical land area, tropical disease and landlockedness seem
to depress economic growth.
there is no support anymore for the finding that «aid is
more efficient in a good policy environment».
In the same logic, Hansen and Tarp (2000) indicate the
excessive sensitivity of Burnside and Dollar's conclusions. Actually, the
significance of the «aid×policy» term depends only on five
observations excluded deliberately from the sample34. As these five
«outliers» are re-included in the database, the conclusion of more
efficient aid in presence of sound economic policy is not valid anymore.
Moreover, Dalgaard and Hansen (2001) show that there is little logical basis
for choosing these particular outliers over other observations. On the basis of
predetermined criteria these five observations would probably have qualified.
Then, they demonstrate how the exclusion of five alternative outliers may
produce a regression that shows a positive impact of aid on growth. Once again
Burnside and Dollar's (1997) conclusions appear to be seriously weakened.
7.5. Problem with the definition of aid
The definition of aid chosen by Burnside and Dollar (1997) is
the «Effective Development Assistance». This original concept
involves only the grant element of aid and excludes for instance, the loan
component of concessional loans. In fact, the most usual definition of aid
comes from the Development Assistance Committee of the OECD and is called
«Official Development Assistance» (ODA). This second concept includes
grants and preferential loans net of repayments of earlier aid loans. For
example, this second definition of aid considers debt cancellation as effective
aid to development whereas the first one does not. Both approaches make sense
and their correlation is obviously high. But when using the alternative
measure, Easterly (2003) find the crucial «aid×policy»
interaction term not to be significant anymore. In the same vein, to test the
robustness of Burnside and Dollar's (1997) findings, Ram (2004) splits the aid
variable into multilateral and bilateral flows. This distinction is certainly
justified, since their allocation procedures are significantly different. He
also tries an alternative concept of policy. All this leads him to reject the
hypothesis of aid effectiveness conditioned by sound economic policy.
34 They are: Nicaragua (1986-9, 1990-3), Gambia (1986-9, 1990-3)
and Guyana (1990-3).
More fundamentally, most studies contain two weaknesses when
they deal with the aid-growth relationship. First, they usually analyse the
relationship between total aid and growth whereas an important part of this aid
is not intended to promote growth. As we already mentioned, no more than thirty
percent of global aid flows are allocated to productive investments. For the
rest, food assistance and humanitarian assistance are not likely to have any
positive impact on growth. As pointed out by Morrissey (2005), the second
limitation has to do with time considerations. Most authors using panel data
for cross-country analysis analyse the aid-growth relationship over periods of
four years. As pointed out by Clemens et al. (2004), this is a very short
period for such an analysis. Financing health or education may only influence
growth over more than a decade. Nevertheless, the longer the period of
observation, the more difficult it is to isolate the specific influence of
aid.
To overcome this dilemma, Clemens et al. (2004) make the
distinction between three different kinds of aid flows following their expected
impact on economic growth. If aid is considered globally, it is logical to find
a small relation with growth. On the other hand, if we restrict the analysis to
the category of aid that is plausible to enhance growth in the short run, then
the impact appears to be more than two times larger than earlier, even for a
four years period of observation35.
Finally, there may be another explanation for the low return
of aid to development. This issue is strangely little discussed in the
scientific literature but recently exposed by some NGOs36. The major
NGO ActionAid International argues that about two third of funds
devoted to international cooperation is actually «phantom aid» that
never reaches its target37. This can be overpriced and ineffective
technical cooperation, tied aid, debt relief, administrative costs or budget
for the hosting of refugees. Actually, donor countries try to present the
largest cooperation budget as possible. But the presence of some particular
amounts within the official budget of development assistance is highly
discussable. Though they recognize the existence of «phantom aid»,
the amount computed by ActionAid International has been largely
contested by the OECD and other bilateral aid agencies. We will not enter
further into this debate. In any case, an important percentage of Official
Development Assistance is lost,
35 Unfortunately, Clemens et al. (2004) do not include an
«aid×policy» interaction term in their regression. For that
reason, the comparison with Burnside and Dollar (1997) is limited.
36 See ActionAid International, June 2005, «Real aid: an
agenda for making aid work».
37 The target is considered to be the improvement of the living
conditions of poor people.
wasted or diverted from its objectives. This has of course
important consequences on the macroeconomic assessment of aid effectiveness.
| 
