7.2. Risky interpretation and sensitivity to model
specification
First, Lensink and White (2000) and Guillaumont (1999)
indicate that the «aid×policy» interaction term needs to be
interpreted carefully. It can mean both that the impact of aid on growth
appreciates with the quality of policy, and that the impact of policy on growth
increases with the quantity of aid.
29 This indicator has been introduced by Sachs and Warner
(1995)
30 CPIA = Country Policy and Institutional Assessment
31 This policy index has been criticised as well. See Dalgaard,
Hansen, and Tarp (2004) for three pertinent remarks.
Furthermore, some authors demonstrated the high sensitivity of
Burnside and Dollar's results to model specification. Among those, Hansen and
Tarp (2001) and Dalgaard and Hansen (2001) modify Burnside and Dollar (1997)
regressions in different ways. For example, when they add an
«aid²» term to the equation, the
«aid×policy» interaction term looses its significance. On the
other hand, aid appears to be effective on average, independently from policy
variables. They also observe diminishing returns.
For Beynon (2001, p29): «There remain significant
unexplained determinants of growth in all these models.» Hansen and Tarp
(2001) also argue that there probably exists other unobserved country-level
effects that give incorrect explanatory power to the
«aid×policy» term. Indeed, if the regression fails to catch
fully and properly all the determinants and constraints that influence economic
growth, the estimated coefficient of aid will be biased. In this sense,
Morrissey (2005, p1) underlines: «As aid is more likely to flow to poor
countries that suffer growth-retarding characteristics (that are not
specified), there is a greater likelihood of incorrectly drawing the conclusion
that aid is ineffective.»
7.3. Other conditioning variables than economic
policy
The literature contains a large number of variables, other
than economic policy, that are presented to have a significant impact on
economic growth. In consequence, there is a high attendant risk that any
individual model does suffer from omitted variable bias and inconsistent
estimators. As Gunning (2004, p54) says: «The idea is that the effect of
aid on growth is conditional on a wide variety of country characteristics, not
just on the policies pursued.» We will briefly present the major
contributions to the discussion.
For Guillaumont and Chauvet (2001), aid is more effective in
economically vulnerable countries32. In those countries, exposed
to external shocks, it appears to be more difficult to maintain consistent
economic policies. They conclude that aid should be allocated in priority to
countries suffering from external shocks, terms of trade difficulties or
natural disasters in order to help them to stabilize. Suspend their assistance
because of poor policies would be very damaging. In the same sense, Collier and
Dehn (2001) say that aid has more impact on growth in countries suffering from
extreme fall of export prices.
32 Economical vulnerability can be measured by the instability of
the agricultural production, the instability of exports earnings, long-term
terms of trade trend and the size of the population.
Bloom and Sachs (1998) and Gallup et al. (1999) all find that
geography has a significant impact on growth33. From this, Dalgaard
et al. (2004) investigate the aid-growth relationship for countries with part
of their territory in the tropical areas. An advantage of this
variable is the absence of endogenity worries. On average, aid seems to
influence positively growth outside the tropics but not in them. In
consequence, Dalgaard et al. (2004) see tropical area as an exogenous
«deep determinant» of growth. If we refer to the principle of
selectivity of Burnside and Dollar (1997), it would obviously be particularly
unfair to penalise tropical countries because of their geographic situation.
The literature contains an extensive variety of other
variables that have been found to condition significantly the efficiency of
aid. Hence, Svensson (1999) highlights that aid is more effective in places
with democratic institutions, whereas Islam (2003) pretends exactly
the opposite, namely totalitarian governments reinforce the impact of
aid on growth. For Petterson (2004), the degree of fungibility of aid
in the recipient country is decisive. Collier and Hoeffler (2002) find out that
countries emerging from conflict have larger absorptive capacities.
All these variables plausibly influence the aid-growth
relationship and there probably exist others. However, Roodman (2003) find that
for most of these studies, the significance of the results were very sensitive
to observations and extensions of dataset. In any case, the conditional
efficiency of aid seems to be much more complex than suggested by Burnside and
Dollar (1997).
| 
