WORKING PAPER NUMBER 32
M
A Y
2007
The Anarchy of Numbers:
Aid, Development, and Cross-country Empirics
By David Roodman
Abstract
Recent literature contains many stories of how foreign aid affects economic growth: aid raises
growth in countries with good policies, or in countries with difficult economic environments, or mainly
outside the tropics, or on average with diminishing returns. The diversity of these results suggests that
many are fragile. I test 7 important aid-growth papers for robustness. The 14 tests are minimally arbi-
trary, deriving mainly from differences among the studies themselves. This approach investigates the
importance of potentially arbitrary specification choices while minimizing arbitrariness in testing
choices. All of the results appear fragile, especially to sample expansion.
The Anarchy of Numbers:
Aid, Development, and Cross-country Empirics
David Roodman1
Research Associate, Center for Global Development
Center for Global Development
Working Paper #32
Final version: May 2007
Abstract: Recent literature contains many stories of how foreign aid affects economic growth: aid raises
growth in countries with good policies, or in countries with difficult economic environments, or mainly
outside the tropics, or on average with diminishing returns. The diversity of these results suggests that
many are fragile. I test 7 important aid-growth papers for robustness. The 14 tests are minimally arbitrary,
deriving mainly from differences among the studies themselves. This approach investigates the importance of potentially arbitrary specification choices while minimizing arbitrariness in testing choices. All
of the results appear fragile, especially to sample expansion.
Keywords: Foreign assistance; economic growth; economic development; robustness testing
JEL codes: F350, O230, O400
1
The author thanks William Easterly and Michael Clemens for advice, Craig Burnside, Lisa Chauvet, Jan Dehn, and
Henrik Hansen for data and assistance, and Stephen Knack, Patrick Guillaumont, anonymous reviewers, and participants in an August 2003 seminar at the Center for Global Development for valuable comments. Responsibility for
the final product rests with the author. Address for correspondence: David Roodman, Center for Global Development, e-mail: droodman@cgdev.org. This paper has appeared in the World Bank Economic Review 21(2), pp. 255–
77, May 2007. The final version differs from the July 2004 one in two major respects. It no longer attempts to correct specification problems such as autocorrelation, focusing purely on the issue of fragility. And it now tests GMM
regressions for robustness to outlier removal, by picking outliers from parallel 2SLS regressions. These changes in
method explain the changes in results since the previous version.
In early 1981, economist Edward Leamer gave a speech at the University of Toronto in which he
bemoaned the state of econometrics. Econometrics sought the status of a science, with regressions its analogue for the reproducible experiments of chemistry or physics. Yet an essential part
of econometric “experimentation” was too often arbitrary, opaque, and unrepeatable. “The
econometric art as it is practiced at the computer terminal involves fitting many, perhaps thousands, of statistical models….This search for a model is often well intentioned, but there can be
no doubt that such a specification search invalidates the traditional theories of inference”
(Leamer 1983). The way out of the quagmire, he argued, was for econometricians to explore larger regions of “specification space,” systematically analyzing the relationship between assumptions and conclusions.
One econometric debate with hallmarks of the syndrome Leamer describes is that on the
effectiveness of foreign aid in developing countries. Since Griffin and Enos (1970), econometricians have parried over the question of how aid affects economic growth in receiving countries.
Prominent in the contemporary work, Burnside and Dollar (2000) conclude, “aid has a positive
effect on growth in a good policy environment.” Their evidence: the statistical significance in
cross-country panel growth regressions of an interaction term of total aid received and an indicator of the quality of recipient economic policies (aid×policy). But Burnside and Dollar is just one
voice among many. Collier and Dehn (2001), Collier and Dollar (2002, 2004), and Collier and
Hoeffler (2004), corroborated their finding, while others challenged it. From the ongoing debate
emerge several stories of the relationship between aid and growth, each of which turns on a particular quadratic or interaction term involving aid. The stories are not incompatible, but most papers support only one. Hansen and Tarp (2001) find that entering the square of aid drives out
aid×policy, and makes the simple aid term significant too: aid works on average, but with dimin-
1
ishing returns. Guillaumont and Chauvet (2001) also fail to find significance for aid×policy, and
instead offer evidence that aid works best in countries with difficult economic environments,
characterized by volatile and declining terms of trade, low population, and natural disasters. In
the same vein, Collier and Dehn (2001) find that increasing aid cushions countries against negative export price shocks. Collier and Hoeffler (2004) offer a triple-interaction term: aid works
particularly well in countries that are recovering from civil war and that have good policies. Last,
Hansen and Tarp, with Dalgaard, say that aid raises growth outside the tropics but not in them
(Dalgaard, Hansen, and Tarp 2004).
These papers differ not only in their conclusions but their specifications too. Within the
group, there are two different choices of period length in the panel data sets, three definitions of
“policy,” three of aid, and four choices of control variable set. Though probably none of the
choices is made on a whim, these differences appear to be examples of what Leamer called
“whimsy.” From Leamer’s point of view, the studies taken together represent a small sampling
of specification space. And few include much robustness testing. Without further analysis, it is
hard to know whether the results reveal solid underlying regularities in the data or are fragile artifacts of particular specification choices.
This paper examines the possibility of fragility systematically. Since by the laws of
chance any regression can be broken with enough experimentation, it is essential for credibility
that the testing suite itself be minimally arbitrary. The tests derive from two sources: the various
choices already present in the original specifications; and the passage of time, which allows expansion of data sets (as in Easterly, Levine, and Roodman 2004). In all, regressions from 7 of the
most prominent studies are subjected to this systematic test suite.
2
Section 1 reviews the approaches and conclusions of the studies that are tested for robustness. Section 2 describes the tests. Section 3 reports the results. Section 4 concludes.
I. History
The hope has often arisen that a turn to the numbers would shed light on the questions of whether
and when foreign aid works. In the view of Hansen and Tarp (2000), this literature has gone
through three generations. The first generation essentially spans 1970–72, and mainly investigates the aid-savings link. Influenced by the Harrod-Domar model, in which savings is the binding constraint on growth, aid-induced saving is assumed to lead directly to investment, thence to
growth via a fixed incremental capital-output ratio. The second generation runs from the early
1970s to the early 1990s and directly investigates whether aid affects investment and growth.
Hansen and Tarp argue that the preponderance of the evidence from these first two generations shows that 1) aid increases total savings, but less than one-to-one; and 2) that aid increases investment and growth. They suggest that studies with more pessimistic results, such as
Mosley, Hudson, and Horrell (1987), have gained disproportionate attention precisely because
they are contrarian.
The third generation commences with Boone (1994) and continues to this day; it is the
focus of this paper. The current generation has brought several innovations. The data sets cover
more countries and years. Reflecting the influence of the new growth theory, regressors are typically included to represent the economic and institutional environment (sometimes together
called the “policy environment”). The potential endogeneity of aid is addressed through instrumenting. And the marginal aid-growth slope is allowed to vary, through incorporation of such
regressors as aid2 and aid×policy. The data sets are almost always panels.
Burnside and Dollar (2000) test whether an interaction term of aid and an index of recipient economic policies is significantly associated with growth. Their panel is drawn from devel3
oping countries outside the former Eastern bloc, covering the six four-year periods in 1970–93.
They incorporate some controls found significant in the general growth literature, namely: initial
income (log real GDP/capita) to capture convergence; ethno-linguistic fractionalization (Easterly
and Levine 1997), assassinations/capita, and the product thereof; the Knack-Keefer (1995) institutional quality variable, “ICRGE”; M2/GDP, to indicate financial depth, lagged one period to
avoid endogeneity (King and Levine 1993); and dummies for sub-Saharan Africa and fastgrowing East Asia.
Burnside and Dollar use a measure of aid called Effective Development Assistance
(EDA, Chang, Fernandez-Arias, and Serven 1998). EDA differs in two major respects from the
usual net Overseas Development Assistance measure (net ODA) tabulated by the Development
Assistance Committee (DAC). First, EDA excludes technical assistance, on the grounds that it
funds not so much recipient governments as consultants. Second, it differs in its treatment of
loans. Net ODA counts disbursements of concessional (low-interest) loans only, but at full face
value.1 As a capital flow concept, it nets out principal but not interest payments on old loans. In
contrast, EDA includes development loans, regardless how concessional (for example, nearcommercial loans by the World Bank to middle-income countries such as Brazil), but counts
only their grant element—that is, their net present value.
Concerned about limited statistical power, Burnside and Dollar combine their economic
policy indicators into a single variable. They first run a growth regression without aid terms, but
with all controls and three indicators of economic policy—log (1+inflation), budget balance/GDP, and the Sachs-Warner (1995) openness variable. All three policy variables differ
1
DAC considers a loan concessional if it has a grant element of at least 25% of the loan value, using a 10% discount
rate.
4
from 0 at the 0.05 level, so Burnside and Dollar form a linear combination of the three using
their coefficients as weights.2
When Burnside and Dollar run their base specification, including aid and aid×policy the
term of central interest, aid×policy, does not in fact enter significantly. However, they find that it
becomes significant after either of two possible changes. Five outlier observations can be excluded (giving Burnside and Dollar’s preferred specification). Or a quadratic interaction term can
be added—aid2×policy, in which case both aid×policy and aid2×policy appear significantly different from 0, the first with positive sign, the second negative. Burnside and Dollar famously
conclude that aid raises growth in a good policy environment, but with diminishing returns.
Burnside and Dollar’s work has triggered responses, some critical, some supportive. Hansen and Tarp (2001) make one prominent attack. They modify the Burnside and Dollar two-stage
least-squares (2SLS) regressions in several ways, most importantly by adding aid2. Aid×policy is
not significant in their results, but aid and aid2 are, the first positive and the second negative. The
implication is that aid is effective on average, but with diminishing returns—regardless of recipients’ policies as far as the evidence goes. Hansen and Tarp then criticize both the Burnside and
Dollar regressions and their own for failing to handle several standard concerns. There may be
country-level fixed effects that correlate with both policies and growth. Failing to purge or control for all such effects could give spurious explanatory power to policies and aid×policy. Also,
variables other than aid and its interaction terms, such as fiscal balance, could be endogenous
and need instrumenting too. They deploy the Arellano-Bond (1991) Generalized Method of
Moments (GMM) estimator, which is designed to handle these problems in short panels. Hansen
2
They also add a constant term to the index, but this has no effect on the regression results of interest here.
5
and Tarp also add Δaid and Δ(aid2) as regressors.3 Their results on aid and aid2 hold. And Δaid
and Δ(aid2) are significant too, again, the first with positive sign and the second negative.
Guillaumont and Chauvet (2001) tell a third story. They hypothesize that the economic
vulnerability of a country influences aid effectiveness. They call economic vulnerability the “environment,” not to be confused with Burnside and Dollar’s “policy environment.” In this story,
aid flows stabilize countries that are particularly buffeted by terms of trade difficulties, other
sorts of external shocks, or natural disasters. Guillaumont and Chauvet build an environment index out of four variables: volatility of agricultural value added (to proxy for natural disasters),
volatility of export earnings, long-term terms of trade trend, and log of population (small countries being more vulnerable to external forces). Their specification is distinctive in using 12-year
periods, and in its controls, which include population growth, mean years of secondary school
education among adults, the Barro-Lee (2000) measure of political instability based on assassinations and revolutions, ethno-linguistic fractionalization, and lagged M2/GDP. In their OLS and
2SLS regressions, aid×environment appears with the predicted negative sign, indicating that aid
works better in countries with worse environments. The term also drives out aid×policy.
Collier and Dollar (2002) corroborate Burnside and Dollar with a quite different data set
and specification. Unlike Burnside and Dollar, they perform OLS only. They include former
Eastern bloc countries, the Bahamas, and Singapore. They use net ODA rather than EDA. They
study 1974–97 instead of 1970–93. They drop all Burnside and Dollar controls except log initial
GDP/capita, ICRGE, and period dummies. But they add region dummies.4 And they define policy as the overall score from the World Bank’s Country Policy and Institutional Assessment
3
This is equivalent to adding lagged aid and lagged aid2 since the regressions also control for aid and aid2.
The regions are Europe and Central Asia, Middle East and North Africa, Southern Asia, East Asia and Pacific,
Sub-Saharan Africa, and Latin America and the Caribbean, as defined by the World Bank.
4
6
(CPIA), which is a composite rating of countries on 20 aspects of policies and institutions.5 Thye
add aid2 but then drop the linear aid term from their preferred specification as insignificant. After
all the changes, aid×policy is again significant, as is aid2, with a negative sign.
Stargin from the Collier and Dollar core regression, Collier and Hoeffler (2004) analyze
how recent emergence from civil war influences aid effectiveness. Sticking to the four-year
panel, they create three dummies to indicate how recently civil war ended. “Peace onset” is 1 in
the period when a country goes from civil war to peace. “Post-conflict 1” is 1 the following period, and “post-conflict 2” the period after that—assuming civil war does not recur.
Aid×policy×post-conflict 1 is significant in Collier and Hoeffler’s preferred (OLS) specification:
aid works particularly well in a good policy environment a few years after civil conflict.
Also corroborating Burnside and Dollar, Collier and Dehn (2001) hew closely to the
Burnside and Dollar specification and data set, and tell a story that incorporates elements from
Guillaumont and Chauvet. They find that adding variables incorporating information on export
shocks renders Burnside and Dollar’s preferred specification—the one with aid×policy but not
aid2×policy—more robust to the inclusion of Burnside and Dollar’s five outliers. First, they add
two variables indicating the magnitude of any positive or negative commodity export price
shocks. They report that aid×policy is then significant at 0.01 for a regression on the full data set.
The negative-shock variable is significant too, with the expected minus sign.6 Then Collier and
Dehn add four aid-shock interaction terms: lagged aid×positive shock, lagged aid×negative
shock, Δaid×positive shock, and Δaid×negative shock. The first and last prove positive and sig-
5
Collier and Hoeffler (2002) make a small correction to the Collier and Dollar data set, excluding 5 observations
where a missing value had been treated as 0. The Collier and Hoeffler version of the Collier and Dollar regression is
tested here.
6
However, the reproduction using their data gives a t statistic of only 0.42 to aid×policy despite having the same R2
and sample size, so their result may be an error. But the same, negative sign does appear on the negative-shock variable.
7
nificant in OLS, and the last, Δaid×negative shock proves particularly robust in their testing. The
study buttresses Burnside and Dollar while suggesting that well-timed aid increases ameliorate
negative export shocks. This matches the Guillaumont and Chauvet result in spirit. But where
Guillaumont and Chauvet interact the amount of aid with the standard deviation of an index of
export volume and other variables, Collier and Dehn’s significant term involves the change in aid
and the change in export prices.
Dalgaard, Hansen, and Tarp (2004) tell a novel aid-growth story. They focus on the share
of a country’s area that is in the tropics, as a determinant of both growth and the influence of aid
on growth. This variable surfaces as a growth determinant in Bloom and Sachs (1998), Gallup
and Sachs (1999), and Sachs (2001, 2003). The causal links may include institutions and economic policies (Acemoglu, Johnson, and Robinson 2001; Easterly and Levine 2003). Dalgaard,
Hansen, and Tarp thus see tropical area as an exogenous “deep determinant” of growth. In the
regressions, aid and aid×tropical area fraction are quite significant, the first with positive sign,
the second with negative sign and similar magnitude. For countries situated completely in the
tropics, the derivative of growth with respect to aid (the sum of the two coefficients) is indistinguishable from 0. Thus, on average, aid seems to work outside the tropics but not in them. The
authors report that their new interaction term drives out both aid×policy and aid2.
There are other third-generation studies (Hadjimichael et al. 1995; Durbarry, Gemmell,
and Greenaway 1998; Lensink and White 2001; Svensson 1999; Chauvet and Guillaumont 2002;
Burnside and Dollar 2004). This paper focuses on those already highlighted as being among the
most influential and, with one exception, having been published. The exception is Collier and
Dehn (2001) which is a pillar of the published Collier and Dollar (2004).
8
The testing here applies to what appear to be authors’ preferred regressions. (See Table
1.) Country by country, the tested regressions generate a diversity of conclusions about the slope
of growth with respect to aid at the margin. (See Table 2, illustrating for the 20 largest aid recipients in 1998.) As an example of the calculations here, the Burnside and Dollar structural equation is
ΔY = αA + β A × P + γP + xδ + ε ,
where Y is GDP/capita, A is aid, P is policy, x is a vector of controls, including initial
GDP/capita, and ε is the error term. So the implied slope of growth respect to aid is
d (ΔY ) dA = α + βP,
which depends on the recipient’s policy level. Applying such formulas to 1998 data, the Burnside and Dollar regression generally predicts benefits from increasing aid while, at the other extreme, the Dalgaard, Hansen, and Tarp and Guillaumont and Chauvet regressions express pessimism. The question is what to make of such conclusions.
II. The Test Suite
There is some robustness testing in the recent literature on aid-growth connections, albeit focusing on Burnside and Dollar. Lu and Ram (2001) introduce fixed effects into the Burnside and
Dollar regressions. Ram (2004) splits the aid variable into the components coming from bilateral
and multilateral donors, and also tests alternative definitions of policy. Dalgaard and Hansen
(2001) modify the choice of excluded outliers. Easterly, Levine, and Roodman (2004) extend the
Burnside and Dollar data set to additional countries and an additional period, 1994–97. All these
tests eliminate the key Burnside and Dollar result. The present study expands Easterly, Levine,
and Roodman along two dimensions. It applies more tests. And it tests more studies.
9
In addition to fragility, the other bugaboo of econometrics is misspecification. Important
questions can be raised about the validity of the regressions tested here. Some exhibit serial correlation in the errors.7 The excludability and relevance of instruments are a legitimate concern.
Regressors treated as exogenous may not be. And term pairs such as aid and aid2 may be multicollinear. But for the sake of concision, this paper focuses on the problem of fragility.
The Tests
The tests applied to these third-generation aid-growth regressions constitute a more systematic
sampling of “specification space” than has hitherto been made. To limit complexity and minimize arbitrariness, each test ideally involves changing just one aspect of the estimations at a
time. The tests are summarized in Table 3. The first four groups of tests, relating to the controls,
the definition of aid and policy, and period length, transfer one specification’s choices to the others.8 Last are tests that modify the sample by dropping outliers and/or expanding to new countries and periods.
The tests are:
1. Changing the control set. In his worries over whimsy, the specification choice that conerns
Leamer is that of regressors. The studies examined use four different control sets, which give
rise to four robustness tests, detailed in Table 3. Each substitutes an alternative control set for
the original one and examines the effect on the significance of key terms.
Like the authors of the original regressions, and in the spirit of avoiding arbitrariness, the
robustness tests here use all complete observations available for developing countries (in7
An earlier version of this paper attempted to address autocorrelation by further modifying the tested specifications,
at the expense of complexity in presentation, and, arguably, “whimsy.” In particular, most of the test regressions
included the log of population as a control since Sargan-type tests suggested it was an improperly excluded instrument. This explains the difference in results between this and the earlier version.
8
Those papers that instrument the variables of interest also differ in their choice of instruments. But since different
variables (aid×policy in one regression, say, versus aid2 in another), ought to be instrumented differently, the various
instrument sets are less interchangeable and of less use for the approach in this paper.
10
cluding the countries of Eastern Europe). Because different variables are available for different subsets of countries, changing the regressor set changes the regression sample. One could
perform variants of the tests that are restricted to the intersections of the old and new samples
in an attempt to distinguish the effects of changing sample and changing variables. This
course is not taken here because it would add to the complexity, would still cause sample
changes, and would not answer the hypothetical, “What would the results have been if the
original authors had used alternative controls?” The authors almost certainly would have
used all available observations.
2. Redefining aid. All the studies take total aid received as a share of recipient GDP. But there
are differences in defining both the numerator and denominator of the ratio. Burnside and
Dollar, Collier and Dehn, and Dalgaard, Hansen, and Tarp use Effective Development Assistance in the numerator while the rest use net ODA. On the choice of denominator, there is
also a split. Hansen and Tarp and Guillaumont and Chauvet use GDP converted to dollars using market exchange rates. The others use real GDP from the Penn World Tables. A country’s relative price level strongly correlates with income per head, with the poorest countries
having price levels 20–25% of that of the United States. Thus using purchasing power parities instead of exchange rates will cause the GDPs of the poorest countries to be measured as
relatively larger and aid to them as relatively smaller as a share of GDP. This might have a
significant effect on coefficient estimates for aid and its interactions
With two options each for measuring aid and GDP, there are four possible combinations
for aid/GDP. The literature includes all but EDA/exchange rate GDP, and these are the bases
for three tests.9 In fact, EDA/real GDP and ODA/real GDP are highly correlated (Dalgaard
9
The published EDA data (Chang, Fernandez-Arias, and Serven 1998) cover only 1975–95. EDA as used here is
extrapolated to the rest of 1970–2001 via a regression of EDA on net ODA.
11
and Hansen 2001), so switching from one to the other may not stress results much. (See
Table 4.)
3.
Redefining good policy. Three sets of “good policy” variables appear among the tested regressions: 1) Burnside and Dollar’s combination of budget balance, inflation, and SachsWarner openness; 2) inflation and Sachs-Warner only (Hansen and Tarp GMM); and 3)
CPIA alone (Collier and Dollar, Collier and Hoeffler). These generate three robustness tests.
Using Burnside and Dollar’s coefficients to form policy indexes (6.85 for budget balance, –
1.40 for inflation, and 2.16 for Sachs-Warner), the first two policy definitions are correlated
0.98, but the third varies more distinctly. (See Table 4.) But in actually application of the
tests, the Burnside-Dollar–style index-forming regression is rerun each time; it includes all
regressors except aid and its interaction terms, and the coefficients on the policy variables are
used to make the index, regardless of statistical significance.10
4. Changing periodization. All but Guillaumont and Chauvet use four-year periods. The lack of
higher-frequency observations of the Guillaumont and Chauvet environment variable prevents adapting their 12-year regressions to a 4-year-period panel. But the other regressions
can be tested on 12-year panels. Notably, key cross-section studies in the growth literature
use periods of 10–25 years despite the small samples that result (Barro 1991; Mankiw, Romer, and Weil 1992; Sachs and Warner 1995).
5. Removing outliers. The tested Burnside and Dollar specification excludes five observations
that are a) outliers in aid×policy and b) highly influential on the coefficient on that term. This
raises a general question about the importance of outliers. To investigate, one robustness test
reruns the reproductions of the original regressions after excluding outliers. Another does the
10
The constant term in the policy index is computed in the same manner as in Burnside and Dollar. It is the predicted growth rate in the model when the policy variables and the period dummies are zero, and all other variables
12
same for the expanded-sample versions. (See below.) Following Easterly, Levine, and
Roodman (2004), outliers are chosen by applying the Hadi (1992) procedure for identifying
multiple outliers to the partial scatter of growth and a regressor of interest, using 0.05 as the
cut-off significance level.11 In 2SLS estimatations, regressors are first projected onto instruments.12
Outliers are not synonymous with influential observations. But even outliers that do not
greatly influence coefficients of interest can substantially affect reported standard errors. In
addition, outliers are the observations most likely to signal measurement problems or structural breaks beyond which the core model does not hold—both of which seem better reasons
for exclusion than high influence. That said, outliers do not necessarily signal measurement
problems or structural breaks. This is especially possible when the variable of interest is
highly non-normal, such as the Collier and Dehn export price shock variable. In such cases,
outliers may contain valuable information about the development process under rare circumstances. Since the two-dimensional partial scatter plot is not well-defined for GMM regressions, in those cases, analogous 2SLS regressions are run for the purposes of identifying outliers.
6. Expanding the sample. Easterly, Levine, and Roodman (2004) develop a dataset that extends
that of Burnside and Dollar from 1970–93 to 1970–97 and adds six countries. For the present
study, that data set has been extended to 2001 and improved in other respects. (See Appendix
1.) This allows a net expansion in both years and countries for all but Guillaumont and
take their sample-average values.
11
Applying the Hadi procedure directly to a full, many-dimensioned data set typically identified 20% or more of
observations as outliers.
12
This test is even run on the Burnside and Dollar 5/OLS regressions, from which one set of outliers is already excluded. Regardless of the genesis of these regressions’ results, it is interesting whether they are driven by a few observations in the remaining sample.
13
Chauvet regression, whose 12-year periods and unusual environment variable hinder expansion.
Issues in Interpreting Results
If Leamer’s (1983) extreme bounds analysis is applied to the results of this testing, then a coefficient will be deemed robustly different from 0 only if it is significantly different from 0 in every
test. However, as Sala-I-Martin (1997) argues, this definition of robustness seems extreme. For
example, one could test robustness by averaging together all observations for each global region,
generating samples of some 6 observations. Almost no regression would pass this test. One could
argue that this test would be “unfair,” i.e., too weak to generate meaningful results. But there is
no sharp division between fair and unfair tests. Indeed, in this test suite, the 12-year-period test
destroys every regression it can be applied to. It is not obvious whether the test is too strong or
the regressions too weak. Thus robustness should be a continuous rather than dichotomous concept.
Sala-I-Martin offers his own procedure for assessing robustness. In essence, he estimates
the cumulative distribution function for a coefficient of interest by running a large number of
variants of the regression it comes from. The robustness of a coefficient is then the fraction of
the density that is on one or the other side of zero. The validity of this concept is based on the
assumption, however informal, that the set of regressions actually run is a representative of all
possible variants of the original regression. For the collection of tests assembled here, that assumption is not valid. For example, one important subset of tests, those expanding the sample,
cannot be applied to the Guillaumont and Chauvet regression. It does not seem plausible that the
test results are representative both with and without this important subset of tests.
14
The sampling of specification space that is made here is minimally arbitrary, but cannot
be assumed to be representative of all possible tests. Thus while Leamer’s definition of robustness may be too harsh for this context, Sala-I-Martin’s has its own limitations. This will be true
even if one performs every possible combination of tests in the suite rather than just one at a
time. In the end, it seems that human judgment applied to the full set of results must substitute
for mechanical definitions of robustness. This in turn means there is some value in keeping the
tests few enough for the human mind to embrace.
III. Results
The first step in the testing is to use the authors’ data sets to reproduce their original results (see
columns 1 of Table 5 and Table 6). All the reproductions exhibit the same pattern of results as
the originals and all but one have the same sample size.13 The Burnside and Dollar, Collier and
Dehn, and Hansen and Tarp reproductions are perfect, and the rest are close. Since the purpose
of the paper is to test robustness, the inexact matches are not a concern. If the results from the
tested regressions are robust, they should withstand whatever minor changes in data or specification cause the discrepancies in the reproductions.
Table 5 and Table 6 report results on key terms in all tested regressions.14 Blank cells indicate inapplicable tests. The test involving the definition of aid as EDA/real GDP, for example,
is not applicable to regressions that originally use it. Using 12-year periods does not work for the
Collier and Hoeffler regression, because the definition of their post-conflict 1 variable assumes
13
The Dalgaard, Hansen, and Tarp regression was executed with the DPD for Ox package (Doornik, Arellano, and
Bond 2002). It turns out that an undocumented limitation in this software—incomplete observations that create gaps
in the time series must always be included in the data file rather than deleted—led to a slight mishandling of the
data. The xtabond2 module for Stata (Roodman 2006), used here, does not have this limitation. This explains the
difference in samples.
14
Full results are available upon request.
15
4-year periods. Lack of higher-frequency data for Guillaumont and Chauvet’s environment variable prevents short-period tests. A total of 77 robustness checks are run.15
Results for tests inspired by differences among the original regressions are in Table 5.
The Collier and Hoeffler result on post-conflict 1×aid×policy (or the collinear post-conflict
1×aid) and the Dalgaard, Hansen, and Tarp results for aid and aid×tropical area fraction do best.
Interestingly, all of these center on sharply bimodal variables: The Collier and Hoeffler postconflict 1 dummy is 1 for only 13 of the 344 observations in their original sample, and there are
negative shocks in 38 of the 234 Collier and Dehn observations. In the Dalgaard, Hansen, and
Tarp sample, 233 of the 371 observations are 100% tropical and 68 are 0%, leaving 70 in between. Evidently regularities involving such variables are more resilient to specification changes.
Results from sample-modifying tests appear in Table 6. The first two result columns are
based on regressions on the original authors’ datasets—first for their full sample, second for the
sample excluding outliers. The next pair of columns is analogous, for the expanded data set. The
figures in Appendix 2 illustrate the sample-modifying results, and are reminders of the importance of checking for outliers. Except for Guillaumont and Chauvet, all the original OLS and
2SLS results depend on outliers for some or all of their significance. The dependence is particularly heavy for the regressions involving aid×policy. On the other hand, the lack of significance
of most of the coefficients under the sample-expansion test is not driven by outliers. It is worth
noting that the Collier and Dehn result on Δaid×negative shock, another interaction term involving a variable with a highly non-normal distribution, is arguably stronger than it looks. The coef-
15
Initial testing revealed multicollinearity in the Collier and Hoeffler regression. In their preliminary regression 3.1
(not regression 3.4, which is tested here), they include the variables post-conflict 1, post-conflict 1×policy, and postconflict 1×aid2, along with the favored post-conflict 1×aid ×policy. In the reproduction of 3.1, post-conflict
1×aid×policy and post-conflict 1×aid have a partial correlation of 0.985, making the two statistically indistinguishable. Thus the Collier and Hoeffler results ought to be interpreted as pertaining to either post-conflict 1×aid ×policy
or post-conflict 1×aid. Occam’s razor argues for the latter.
16
ficient is reversed by the exclusion of outliers from the original sample. But it is arguably fallacious to draw conclusions about the role of shocks having excluded many of the most dramatic
examples.
The overall pattern is clear-cut. The 12-year test is the toughest, probably because of the
small samples, failing all regressions. The new-data test is not far behind, an important point
given that the surgery it involves—a moderate sample expansion—is much less radical. Reading
the tables by rows (test subjects) instead of columns (tests), we see that the Dalgaard, Hansen,
and Tarp result on the aid-tropics link is the only one to come through the specificiationmodifying tests strongly. But it too falls down on the sample-modifying tests after outliers are
removed. Four of the nine outliers are for the Jordan, covering 1974–89, a period in which that
non-tropical country experienced high growth and received much aid from its neighbors. This
confirms the conclusion of Rajan and Subramanian (2005) that the aid×tropics result is fragile
too.
IV. Conclusion
Each of the papers examined here embodies a set of choices about model specification and data.
Aid is measured a certain way. A certain epoch is studied. Periods have a certain length. And so
on. Some of these choices imply assumptions about the world, such as, say, that aid is exogenous
to growth. All limit the scope of a strict interpretation of the results. A question of great importance for the literature is, how many of such implied assumptions can be dropped without harming the conclusions?
The results reported here suggest that the fragility found in Easterly, Levine, and Roodman (2004) for Burnside and Dollar is the norm in the cross-country aid effectiveness literature.
Indeed, in a counterpoint to the focus of Leamer (1983), Levine and Renelt (1992), and Sala-I-
17
Martin (1997) on the choice of controls as a source of fragility, it turns out that modifying the
sample generally affects results most. For example, in the Collier and Dollar regression, half of
the specification-modifying tests leave the t statistic at 1.49 or higher and two more lower it to
near 1.00. (See Table 5.) But adding more years sends it to –0.19—and, after dropping outliers,
to –0.81. (See Table 6.)
Does this mean that the various stories of aid effectiveness should be summarily dismissed? Are recipient policies, exogenous economic factors, and post-conflict status irrelevant to
aid effectiveness? Are there no diminishing returns to aid? Is helping the neediest countries a
hopeless task? No. There can be no doubt that some aid finances investment, and that domestic
policies, governance, external conditions, and other factors these authors study influence the
productivity of investment. Why then do such stories of aid effectiveness not shine through more
clearly? The reasons are several. Aid is probably not a fundamentally decisive factor for development, not as important as, say, domestic savings, inequality, or governance. Moreover, foreign
assistance is not homogeneous. It consists of everything from food aid for famine-struck countries to technical advice on building judiciaries to loans for paving roads. And much aid is poorly
used—or, like venture capital, is good bets gone bad. Thus the statistical noise tends to drown
out the signal.
Perhaps researchers will yet unearth more robust answers to the fundamental questions of
aid policy. Or perhaps they have hit the limits of cross-country empirics. Either way, robust,
valid generalizations have not and will not come easily. Despite decades of trying, cross-country
growth empirics have yet to teach us much about whether and when aid works.
18
Table 1. Regressions Tested
Regression
Estima- Former East
tor
bloc?
Burnside &
Dollar
5/OLS
OLS
No
Collier &
Dehn 3.4
““
““
Collier &
Dollar 1.21
OLS
Yes
Collier &
Hoeffler
3.4
OLS
““
Hansen &
Tarp 3.2
Difference
GMM
Dalgaard,
Hansen, System
and Tarp GMM
3.5
Definition of
Study Years/
Outliers Key significant
Aid
Policy
Controls
period period
out?
term(s)
LGDP, ETHNF,
ASSAS,
BB,
EDA/real
ETHNF×ASSAS, 1970– 4
Yes
INFL,
aid×policy
GDP
ICRGE, M2, SSA, 93
SACW
EASIA, period
dummies
aid×policy,
““
1974–
““
““
““
No
×negative
Δaid
93
shock
LGDP, ICRGE,
1974–
ODA/real
aid×policy,
““
CPIA
““
policy, period and
97
GDP
aid2
region dummies
““
““
““
““
““
““
aid×policy×
post-conflict 1
No
LGDP, ASSAS,
ETHNF×ASSAS, 1978–
93
ICRGE, M2, period dummies
““
ODA/
INFL,
exchange
SACW
rate GDP
““
aid, aid2, Δaid,
Δaid2
Yes
LGDP, policy, pe- 1970–
riod dummies
97
““
EDA/real
GDP2
BB,
INFL,
SACW
““
aid,
aid×tropical
area fraction
GuillauLGDP, ENV, SYR,
aid,
ODA/
mont &
POPG, M2, PIN- 1970–
2SLS
No
12 exchange
““
““
aid×environChauvet
STAB, ETHNF,
93
rate GDP
ment
5.2
period dummy
1
2
As revised in Collier & Hoeffler 1.1. As extrapolated to 1970–74 and 1996–97 in Easterly, Levine, and
Roodman (2004).
Abbreviations: LGDP=log initial real GDP/capita; ETHNF=ethno-linguistic fractionalization, 1960; ASSAS=assassinations/capita; ICRGE=composite of International Country Risk Guide governance indicators; M2=M2/GDP, lagged; SSA=Sub-Saharan Africa dummy; EASIA=fast-growing East Asia dummy;
ENV=Guillaumont & Chauvet “environment” variable; SYR=mean years of secondary schooling among
adults; PINSTAB=average of ASSAS and revolutions/year; BB=budget balance/GDP;
INFL=log(1+inflation); SACW=Sachs-Warner openness; EDA=Effective Development Assistance;
ODA=Net Overseas Development Assistance.
Source: Author’s analysis based on sources described in the text.
19
Table 2. Marginal Impact of Aid According to Preferred Regression, Various Studies, for
20 Largest Aid Recipients of 1998
Country
Bangladesh
Bolivia
China
Cote d'Ivoire
Egypt
Ethiopia
Haiti
India
Indonesia
Kenya
B&D 5-OLS
0.20
(0.15)
0.56
(0.22)
0.16
(0.15)
0.56
(0.22)
0.59
(0.23)
0.30
(0.16)
0.15
(0.15)
0.13
(0.15)
0.61
(0.23)
0.46
(0.19)
Collier &
Dehn
–0.04
(0.15)
0.15
(0.22)
–0.06
(0.14)
0.15
(0.22)
0.16
(0.22)
0.01
(0.16)
–0.07
(0.14)
–0.08
(0.14)
0.17
(0.23)
0.09
(0.19)
0.57
(0.22)
0.61
(0.23)
0.54
(0.21)
–0.03
(0.17)
0.44
(0.19)
0.62
(0.23)
0.50
(0.20)
0.19
(0.15)
0.08
(0.15)
0.64
(0.45)
0.17
(0.23)
0.13
(0.21)
–0.17
(0.13)
0.99
(0.78)
0.18
(0.23)
1.56
(1.24)
–0.05
(0.14)
–0.11
(0.13)
Mozambique
Nicaragua
Philippines
Poland
Russia
Tanzania
Thailand
Uganda
Vietnam
Zambia
Collier &
Dollar
0.43
(0.20)
0.29
(0.16)
0.53
(0.25)
0.11
(0.10)
0.30
(0.15)
0.20
(0.12)
–0.23
(0.13)
0.48
(0.22)
0.51
(0.24)
0.33
(0.16)
0.06
(0.08)
–0.09
(0.12)
0.43
(0.20)
0.42
(0.20)
0.36
(0.17)
0.09
(0.08)
0.57
(0.27)
0.16
(0.11)
0.45
(0.21)
–0.38
(0.21)
Collier &
Hoeffler
0.37
(0.20)
0.24
(0.15)
0.46
(0.24)
0.07
(0.09)
0.25
(0.14)
0.71
(0.15)
–0.25
(0.12)
0.41
(0.22)
0.44
(0.24)
0.28
(0.16)
0.53
(0.12)
0.51
(0.18)
0.37
(0.20)
0.36
(0.20)
0.32
(0.17)
0.05
(0.08)
0.49
(0.26)
0.11
(0.11)
0.39
(0.21)
–0.40
(0.19)
Hansen &
Dalgaard et Guillaumont
Tarp GMM
& Chauvet
al.
0.14
0.29
–0.72
(0.13)
(0.18)
(0.37)
–0.02
–0.29
–0.33
(0.08)
(0.33)
(0.17)
0.20
0.66
(0.15)
(0.14)
–0.05
–0.29
–0.23
(0.08)
(0.33)
(0.13)
0.13
0.53
–0.55
(0.13)
(0.14)
(0.28)
–0.08
–0.29
0.04
(0.08)
(0.33)
(0.12)
–0.28
–0.29
(0.10)
(0.33)
0.20
0.19
–0.97
(0.15)
(0.20)
(0.51)
0.19
–0.29
–0.68
(0.15)
(0.33)
(0.35)
0.06
–0.29
–0.48
(0.10)
(0.33)
(0.25)
–1.04
–0.20
–0.26
(0.39)
(0.30)
(0.14)
–0.65
–0.29
–0.18
(0.24)
(0.33)
(0.11)
0.18
–0.29
–0.86
(0.14)
(0.33)
(0.45)
0.17
0.69
(0.14)
(0.14)
0.20
0.69
(0.15)
(0.14)
–0.21
–0.29
(0.09)
(0.33)
0.20
–0.29
–0.80
(0.15)
(0.33)
(0.42)
–0.09
–0.29
–0.13
(0.08)
(0.33)
(0.10)
0.12
–0.29
0.00
(0.12)
(0.33)
(0.00)
–0.43
–0.29
–0.32
(0.15)
(0.33)
(0.17)
Robust standard errors in parentheses. All figures are based on reproductions of the original regressions. All pertain to
1994-97, except those the the Guillaument & Chauvet regression, which pertain to 1982-93. Aid is taken as a share of
exchange rate GDP in the Hansen & Tarp and Guilaumont & Chauvet regressionis and of PPP GDP in the rest. Blank
cells are causing by missing observations of underlying indicators.
Source: Authors' analysis based on sources described in the text.
20
Table 3. Robustness Tests
Test
Changing controls
BD controls
CD controls
GC controls
DHT controls
Changing aid definition
EDA/real GDP
ODA/real GDP
ODA/exchange rate GDP
Changing policy definition
INFL, BB, SACW
INFL, SACW
CPIA
Description
Control for LGDP, ETHNF, ASSAS, ETHNF×ASSAS, ICRGE, M2, SSA,
EASIA, period effects, as in Burnside & Dollar, Collier & Dehn, Hansen
& Tarp
Control for LGDP, ICRGE, period and region effects, as in Collier &
Dollar, Collier & Hoeffler
Control for LGDP, ENV, SYR, POPG, M2, PINSTAB, ETHNF, period effects, as in Guillaumont & Chauvet
Control for LGDP, ICRGE, SSA, EASIA, period effects, as in Dalgaard,
Hansen, and Tarp
Effective Development Assistance/real GDP, as in Burnside & Dollar,
Collier & Dehn, Dalgaard, Hansen, & Tarp
Net Overseas Development Assistance/real GDP, as in Collier & Dollar,
Collier & Hoeffler
Net Overseas Development Assistance/exchange rate GDP, as in
Hansen & Tarp, Guillaumont & Chauvet
Inflation, budget balance, and Sachs-Warner openness, as in Burnside &
Dollar, Collier & Dehn
Inflation and Sachs-Warner, as in Hansen & Tarp
Country Policy and Institutional Assessment, as in Collier & Dollar, Collier
& Hoeffler
Changing period length
12-year
Aggregate over 12-year periods, as in Guillaumont & Chauvet
Changing sample and data set
No outliers
Remove Hadi outliers in the partial scatter of the dependent variable and
the independent variable of greatest interest
Expanded sample
New data set. Carried to 2001, except shocks data end in 1997 and Guillaumont & Chauvet environment variable not updated
Expanded sample, no outCombine above two changes
liers
Abbreviations as in Table 2.
21
Table 4. Simple Correlations of Aid and Good Policy Measures, Respectively, Four-Year
Periods, on Available Observations
EDA/real GDP
ODA/real GDP
ODA/exchange rate
GDP
EDA/real GDP
1.00
ODA/real GDP
0.97
1.00
ODA/exchange rate GDP
0.78
0.82
1.00
Inflation, budget bal-
Inflation, Sachs-Warner
CPIA
ance, Sachs-Warner
Inflation, budget balance,
Sachs-Warner
1.00
Inflation, Sachs-Warner
0.98
1.00
CPIA
0.53
0.52
1.00
Source: Author’s analysis based on sources described in the text.
22
Table 5. Coefficients on key terms under specification-modifying tests (original data set)
Changing controls
Specification key term Original
Aid ×
0.19
Burnside &
policy
(2.61)
Dollar
Collier &
Dehn
Collier &
Dollar
Collier &
Hoeffler
Aid ×
policy
Δaid ×
negative
shock
Aid ×
policy
Postconflict 1
× aid ×
policy
Aid
2
Aid
Hansen &
Tarp
BD
Aid
0.06
0.20
0.16
0.05
(1.07)
(2.31)
(2.07)
(1.96)
0.29
–0.03
(3.12) (–0.25)
279
263
276
275
275
296
264
0.02
0.05
0.07
0.02
0.12
0.11
(1.70)
(0.47)
(0.43)
(0.84)
(1.19)
(1.39)
(2.02)
(0.83)
0.04
0.02
0.03
0.02
0.04
0.01
0.03
0.02
(3.17)
(1.12)
(1.53)
(1.06)
(2.67)
(4.27)
(2.04)
(0.82)
234
242
227
242
234
234
256
268
0.14
0.12
–0.07
0.12
0.17
0.03
0.05
0.06
(2.15)
(1.84)
(–1.91)
(1.87)
(1.70)
(1.50)
(1.00)
(1.10)
344
337
374
349
349
347
365
388
0.18
0.18
0.13
0.18
0.29
0.04
0.18
0.18
(3.92)
344
(3.89)
337
(2.82)
374
(3.81)
349
(3.75)
349
(4.07)
347
(3.97)
365
(4.23)
388
0.90
1.10
1.00
1.10
–0.36
1.85
0.94
0.69
(4.22)
(1.56)
(2.57)
(1.56)
(–0.25)
(1.30)
(2.02)
(1.53)
–0.02
–0.02
–0.03
–0.02
–0.04
–0.17
–0.02
–0.02
(–1.19) (–2.81) (–1.19)
(–0.18)
(–2.48)
(–1.81)
(–1.76)
–0.47
–0.47
–0.47
–1.46
–0.71
–0.69
(–1.42) (–1.56) (–1.42)
–0.58
(–0.70)
(–1.74)
(–2.63)
(–2.48)
0.01
0.01
0.02
0.01
0.07
0.12
0.01
0.01
(3.64)
(1.10)
(1.66)
(1.10)
(0.53)
(2.24)
(1.86)
(2.11)
213
214
213
215
213
Dalgaard,
Hansen, &
Tarp
0.15
(2.09)
0.03
–0.70
2
DHT
270
(–4.91)
Δ(Aid )
GC
0.10
(–3.83)
ΔAid
CD
Changing policy
Changing aid
EDA/ ODA/ ODA/ INFL, INFL,
real
real
XR
BB, SAC
CPIA
W
GDP GDP GDP SAC
213
181
0.69
1.17
1.34
1.33
1.10
214
0.46
(5.09)
(6.44)
(6.62)
(5.72)
(5.23)
(4.04)
Aid ×
–0.98 –1.49 –1.79 –1.66
–1.17
–0.45
tropical
(–3.16) (–8.58) (–8.26) (–6.72)
(–5.09) (–3.79)
area %
371
354
371
315
371
365
Aid ×
–0.15 –0.13 –0.07
–0.09
–0.45
–0.31
–0.15
–0.13
Guillaumont
environ(–1.79) (–1.73) (–1.40)
(–1.54) (–1.96) (–1.77)
(–2.13) (–2.01)
& Chauvet
ment
68
71
73
73
66
66
68
69
All t statistics (in parentheses) are heteroskedasticity-robust; those for GMM regressions also autocorrelation
Except for original Hansen and Tarp regression, all GMM standard errors incorporate the Windmeijer (2005)
sample correction. Entries significant at 0.05 in bold.
Source: Authors’ analysis based on sources described in the text.
23
Table 6. Coefficients on Key Terms under Data Set–Modifying Tests
Specification
Burnside & Dollar
key term
Aid × policy
Observations
Aid × policy
Collier & Dehn
Collier & Dollar
Collier & Hoeffler
ΔAid × negative shock
Observations
Aid × policy
Observations
Post-conflict ×
aid × policy
Observations
Aid
2
Aid
Hansen & Tarp
Aid, lagged
2
Aid , lagged
Observations
Aid
Dalgaard, Hansen, &
Tarp
Original
Full
No
sample outliers
New data
Full
No
sample outliers
0.19
–0.05
–0.03
–0.25
(2.61)
(–0.45)
(–0.17)
(–1.27)
270
263
446
436
0.10
0.11
0.06
0.10
(1.70)
(1.11)
(1.07)
(0.73)
0.04
–0.06
0.03
–0.17
(3.17)
(–1.33)
(2.93)
(–1.92)
234
224
402
379
0.14
0.07
–0.01
–0.04
(2.15)
(1.06)
(–0.19)
(–0.81)
344
341
520
508
0.18
1.18
0.08
–0.06
(3.92)
(2.12)
(1.91)
(–0.19)
344
333
520
494
0.90
0.96
0.08
0.03
(4.22)
(2.19)
(0.41)
(0.16)
–0.02
–0.02
–0.001
–0.001
(–3.83)
(–1.95)
(–0.57)
(–0.42)
–0.70
–0.70
–0.13
–0.07
(–4.91)
(–2.73)
(–0.92)
(–0.53)
0.01
0.01
0.002
0.001
(3.64)
(1.86)
(1.03)
(0.71)
213
212
517
514
0.69
0.34
0.89
–0.69
(5.09)
(0.20)
(5.23)
(–0.79)
Aid × tropical
–0.41
0.63
–0.98
–0.96
area %
(–0.25)
(0.73)
(–3.16)
(–3.65)
Observations
371
362
474
463
Aid ×
–0.15
–0.11
environment
Guillaumont & Chauvet
(–1.79)
(–1.96)
Observations
68
67
All t statistics (in parentheses) are heteroskedasticity-robust; those for GMM
regressions also autocorrelation-robust. Except for original Hansen and Tarp
regression, all GMM standard errors incorporate the Windmeijer (2005) finitesample correction. Entries significant at 0.05 in bold.
24
References
Acemoglu, D., S. Johnson, and J. A. Robinson. 2001. “The colonial origins of comparative development: An empirical investigation.” American Economic Review 91: 1369–1401.
Anderson, T.W. and C. Hsiao. 1982. “Formulation and Estimation of Dynamic Models Using
Panel Data.” Journal of Econometrics 18(1): 47–82.
Arellano, M. and S. Bond. 1991. “Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations.” The Review of Economic Studies 58(2):
277–97.
Banks, A. 2002. “Cross-National Time-Series Data Archive.” Databanks International, Bronx,
NY.
Barro, R. J. 1991. “Economic Growth in a Cross Section of Countries.” American Economic Review 106(2): 407–43.
Barro, R. J. and J. Lee. 2000. “International Data on Educational Attainment: Updates and Implications.” Working Paper 7911. National Bureau of Economic Research, Cambridge, MA.
Bloom, D. and J. D. Sachs. 1998. “Geography, Demography and Economic Growth in Africa.”
Brookings Papers on Economic Activity 2: 207–73.
Boone, P. 1994. “Aid and Growth.” Mimeo, London School of Economics, London.
Burnside, C. and D. Dollar. 2000. “Aid, Policies, and Growth.” American Economic Review
90(4): 847–68.
Burnside, C. and D. Dollar. 2004. “Aid, Policies, and Growth: Revisiting the Evidence.” Policy
Research Working Paper O-2834. World Bank, Washington, DC.
Chang, C. C., E. Fernandez-Arias, and L. Serven. 1998. “Measuring Aid Flows: A New Approach.” Working Paper 387. Inter-American Development Bank, Washington, DC.
Chauvet, L. and P. Guillaumont. 2002. “Aid and Growth Revisited: Policy, Economic Vulnerability and Political Instability.” Paper presented at the Annual Bank Conference on Development Economics: Towards Pro-poor Policies, Oslo.
Collier, P. and J. Dehn. 2001. “Aid, Shocks, and Growth.” Policy Research Working Paper
2688. World Bank, Washington, DC.
Collier, P. and D. Dollar. 2002. “Aid Allocation and Poverty Reduction.” European Economic
Review 45(1): 1–26.
Collier, P. and D. Dollar. 2004. “Development Effectiveness: What Have We Learnt?” The Economic Journal 114(496): F244–71.
Collier, P. and A. Hoeffler. 2004. “Aid, Policy and Growth in Post-Conflict Societies.” Euro25
pean Economic Review 48(5): 1125–45.
Dalgaard, C. and H., Hansen. 2001. “On Aid, Growth and Good Policies.” Journal of Development Studies 37(6): 17–41.
Dalgaard, C., H. Hansen, and F. Tarp. 2004. “On the Empirics of Foreign Aid and Growth.” The
Economic Journal 114(496): F191–F216.
Dehn, J. 2000. “Commodity Price Uncertainty in Developing Countries.” Working Paper 12.
Centre for the Study of African Economies, Oxford.
Development Assistance Committee (DAC). 2002. Development Assistance Committee Online.
Paris.
Doornik, J.A., M. Arellano, and S. Bond. 2002. “Panel Data Estimation Using DPD for Ox.”
Available online at http://www.doornik.com/download/dpd.pdf.
Durbarry, R., N. Gemmell, and D. Greenaway. 1998. “New Evidence on the Impact of Foreign
Aid on Economic Growth.” CREDIT Research Paper 98r8. University of Nottingham, Nottingham.
Easterly, W. and R. Levine. 1997. “Africa’s Growth Tragedy: Policies and Ethnic Divisions.”
Quarterly Journal of Economics 112(4): 1203–50.
Easterly, W. and R. Levine. 2003. “Tropics, Germs, and Crops: How Endowments Influence
Economic Development.” Journal of Monetary Economics 50(1): 3–39.
Easterly, W., R. Levine, and D. Roodman. 2004. “New Data, New Doubts: A Comment on
Burnside and Dollar’s ‘Aid, Policies, and Growth (2000)’.” American Economic Review
94(2).
Easterly, W. and S. Rebelo. 1993. “Fiscal Policy and Economic Growth: An Empirical Investigation.” Journal of Monetary Economics 32(3): 417–58.
Gallup, J. L., and J. D. Sachs. 1999. “Geography and Economic Development.” In Pleskovic, B.
and Stiglitz, J. E. eds., Annual World Bank Conference on Development Economics, 1998
Proceedings. World Bank, Washington, DC, pp. 127–78.
Griffin, K. B. and J. L. Enos. 1970. “Foreign Assistance: Objectives and Consequences.” Economic Development and Cultural Change 18(3): 313–27.
Guillaumont, P. and L. Chauvet. 2001. “Aid and Performance: A Reassessment.” Journal of Development Studies 37(6): 66–92.
Hadi, A. S. 1992. “Identifying Multiple Outliers in Multivariate Data.” Journal of the Royal Statistical Society, Series B 54: 761-777.
Hadjimichael, M. T., D. Ghura, M., Muhleisen, R. Nord, and E. M. Ucer. 1995. “Sub-Saharan
26
Africa: Growth, Savings, and Investment, 1986–93.” Occasional Paper 118. International
Monetary Fund, Washington, DC.
Hansen, H. and F. Tarp. 2000. “Aid Effectiveness Disputed.” Journal of International Development 12(3): 375–98.
Hansen, H. and F. Tarp. 2001. “Aid and Growth Regressions.” Journal of Development Economics 64(2): 547–70.
Holtz-Eakin, D., W. Newey, and H. S. Rosen. 1988. “Estimating Vector Autoregressions with
Panel Data.” Econometrica 56(6): 1371–95.
International Monetary Fund (IMF). 2003. International Financial Statistics Database. Washington, DC.
Jepma, C. J. 1991. “The Tying of Aid.” OECD Development Centre, Paris.
Kaufmann, D., A. Kraay, and M. Mastruzzi. 2003. “Governance Matters III: Governance Indicators for 1996–2002.” Policy Research Working Paper 3106. World Bank, Washington, DC.
Killick, T. 1998. “Aid and the Political Economy of Policy Change.” Routledge, London.
King, R. G. and R. Levine. 1993. “Finance and Growth: Schumpeter Might be Right.” American
Economic Review 108(3): 717–37.
Knack, S. and P. Keefer. 1995. “Institutions and Economic Performance: Cross-Country Tests
Using Alternative Institutional Measures.” Economics and Politics 7(3): 207–27.
Leamer, E. E. 1983. “Let’s Take the Con out of Econometrics.” American Economic Review
73(1): 31–43.
Levine, R. and D. Renelt. 1992. “A Sensitivity Analysis of Cross-Country Growth Regressions.”
American Economic Review 82(4): 942–63.
Lensink, R. and H. White. 2001. “Are There Negative Returns to Aid?” Journal of Development
Studies 37(6): 42–65.
Lu, S. and R. Ram. 2001. “Foreign Aid, Government Policies, and Economic Growth: Further
Evidence from Cross-country Panel Data for 1970–93.” Economia Internazionale 54(1): 15–
29.
Mankiw, N. G., D. Romer, and D. N. Weil. 1992. “A Contribution to the Empirics of Economic
Growth.” American Economic Review 107(2): 407–37.
Mosley, P., J. Hudson, and S. Horrell. 1987. “Aid, the Public Sector and the Market in Less Developed Countries.” Economic Journal 97(387): 616–41.
Rajan, R. and A. Subramanian. 2005. “Aid and Growth: What Does the Cross-Country Evidence
Really Show?” Working Paper 05/127. International Monetary Fund, Washington, DC.
27
Ram, R. 2004. “Recipient Country’s ‘Policies’ and the Effect of Foreign Aid on Economic
Growth in Developing Countries: Additional Evidence.” Journal of International Development 16(2): 201–11.
Rodríguez, F. and D. Rodrik. 2001. “Trade Policy and Economic Growth: A Skeptic's Guide to
the Cross-National Evidence.” In Bernanke, B. and Rogoff, K. S., eds., National Bureau of
Economic Research Macroeconomics Annual. MIT Press, Cambridge, MA.
Roeder, P.G. 2001. “Ethnolinguistic Fractionalization (ELF) Indices, 1961 and 1985.” Avaliable
on http//:weber.ucsd.edu/~proeder/elf.htm, accessed May 2004.
Roodman, D. 2006. “How to Do xtabond2: An Introduction to “Difference” and “System” GMM
in Stata.” Working Paper 103. Center for Global Development, Washington, DC.
Sachs, J. D. 2001. “Tropical Underdevelopment.” Working Paper W8119. National Bureau of
Economic Research, Cambridge, MA.
Sachs, J. D. 2003. “Institutions Don’t Rule: Direct Effects of Geography on per Capita Income.”
Working Paper W9490. National Bureau of Economic Research, Cambridge, MA.
Sachs, J. D. and A. Warner. 1995. “Economic Reform and the Process of Global Integration.”
Brookings Papers on Economic Activity 1995(1): 1–118.
Sala-I-Martin, X. X. 1997. “I Just Ran Two Million Regressions.” American Economic Review
87(2): 178–83.
Summers, R. and A. Heston. 1991. “The Penn World Table (Mark 5): An Expanded Set of International Comparisons, 1950–88.” Quarterly Journal of Economics 106(2): 327–68.
Svensson, J. 1999. “Aid, Growth and Democracy.” Economics and Politics 11(3): 275–97.
Van de Walle, N. and T. A. Johnston. 1996. “Improving Aid to Africa.” Policy Essay 11. Overseas Development Council, Washington, DC.
United Kingdom Department for International Development (DFID). 2000. “Eliminating World
Poverty: Making Globalisation Work for the Poor.” White Paper on International Development Presented to Parliament by the Secretary of State for International Development by
Command of Her Majesty, London.
United States Department of State (various years). World Military Expenditures and Arms Transfers. Washington, DC.
Wacziarg, R. and K. H. Welch. 2002. “Trade Liberalization and Growth: New Evidence.”
Mimeo, Stanford University, Stanford, CA.
Windmeijer, F. 2005. “A Finite Sample Correction for the Variance of Linear Two-step GMM
Estimators.” Journal of Econometrics 126: 25–51.
28
World Bank. 2003. World Development Indicators 2003 Database. Washington, DC.
29
Appendix 1. Data set construction
The data set used in this study is based on that of Easterly, Levine, and Roodman (2004). Some
variables in that set have been slightly revised. Others have been added to match the data sets of
the tested regressions. The period of coverage has been pushed back to 1958 and forward to 2001
where possible. All data were collected from standard sources, expect for countries’ export price
indexes, which were provided by Jan Dehn (see Dehn 2000). (See Table A1.)
Following are notes on the data set construction:
Revisions since Easterly, Levine, and Roodman (2004)
•
Some observations for inflation are completed by using wholesale inflation where consumer price inflation was unavailable.
•
The update of the Sachs-Warner variable is slightly revised under the influence of the independent update by Wacziarg and Welch (2002).
•
Some missing values for Effective Development Assistance during 1975–95, the period
of the EDA data set, are filled in in the same manner as missing values outside this period
already were, via a regression of EDA on Net ODA.
•
ICRGE now varies over time, rather than taking 1982 values throughout. Observations
before 1982 are assigned 1982 values. In addition, the variable is revised in order to extend it after 1997. In 1998, the PRS Group stopped reporting two of ICRGE’s original
components, Expropriation Risk and Repudiation of Government Contracts. So these
were dropped entirely from the variable, leaving Corruption, Bureaucratic Quality, and
Rule of Law. On annual data, the revised ICRGE has a 0.97 correlation with the original.
•
Some missing values for ethno-linguistic fractionalization are filled in from Roeder
30
(2001).
Expansion of period
•
Data was collected where available for 1958–2001.
•
The Collier and Dehn shocks variables are only updated to 1997 because the underlying
data on export prices in Dehn (2000) cease in 1997.
•
The Guillaumont and Chauvet environment variable is not updated, for lack of underlying data on its four components.
•
The 1998–2001 values for the updated Sachs-Warner variable are based on 1998 data
only. Currency Data International, the long-time source of black market premium data,
which is one component of Sachs-Warner, shut down in 1999.
Table A2 documents the changes in sample that the new data set brings to the tested regressions.
Table A1. Construction of Data Set
Variable
Per-capita GDP
growth
Initial GDP per
capita
Code
GDPG
Data source
World Bank 2003
Notes
LGDP
Summers and Heston
1991, updated using
GDPG
Ethno-linguistic
fractionalization,
1960
Assassinations/
capita
Political instability, lagged
Institutional quality
ETHNF
Roeder 2001
ASSAS
Banks 2002
Natural logarithm of
GDP/capita for first year of
period; constant 1985 dollars
Probability that two randomly chosen individuals
differ ethnically
Assassinations/capita
PINSTAB
Banks 2002
ICRGE
PRS Group’s IRIS III
data set (see Knack
and Keefer 1995)
M2/GDP, lagged
one period
Sub-Saharan Africa
East Asia
M2–1
World Bank 2003
SSA
World Bank 2003
EASIA
Simple average of ASSAS
and revolutions/year
Revised. Computed as the
average of the three components still reported after
1997, dropping two.
Codes nations in the southern Sahara as sub-Saharan
Dummy for China, Indone-
31
sia, South Korea, Malaysia,
Philippines, and Thailand
Central America
Franc zone
CENTAM
FRZ
Egypt
Budget surplus
EGYPT
BB
Inflation
INFL
World Bank 2003; IMF
2003
Sachs-Warner,
updated
SACW
Positive and
negative shock
POSSHOCK
NEGSHOCK
Sachs and Warner
1995; Easterly, Levine, and Roodman
2004; Wacziarg and
Welch 2002
Dehn 2000
Effective Development Assistance/real GDP
AID
Chang, FernandezArias, and Serven
1998; OECD-DAC
2002; IMF 2003;
World Bank 2003;
Summer and Heston
1991
Net Overseas
Development Assistance/real
GDP
Net Overseas
Development Assistance/nominal
GDP
Dummy for end of
civil conflict in
previous period
Tropical area
fraction
Population
Population
growth
Mean years of
secondary
schooling among
ODAPPPGDP
OECD-DAC 2002;
IMF 2003; World Bank
2003; Summer and
Heston, 1991
OECD-DAC 2002;
World Bank 2003
ODAXRGDP
World Bank 2003
Burnside and Dollar
2000
World Bank 2003; IMF
2003
POSTCONFLICT1
Collier and Hoeffler
2004
TROPICAR
LPOP
POPG
Gallup and Sachs
1999
World Bank 2003
World Bank 2003
SYR
Barro and Lee 2000
Codes African nations in
the CFA franc zone
World Bank primary
source. Additional values
extrapolated from IMF, using series 80 and 99b (local-currency budget surplus
and GDP)
log (1 + inflation). World
Bank primary source.
Wholesale price inflation
from IMF used to fill gaps
Extended to 1998. Slightly
revised pre-1993.
Shocks are % price index
changes. “Shock” threshold
country-specific. Reconstructed based on underlying index data for 1957–97
Available values for 1975–
95 from Chang, FernandezArias, and Serven. Missing
values extrapolated based
on regression of EDA on
Net ODA. Converted to
1985 dollars with World
Import Unit Value index
from IMF, series 75. GDP
computed like LGDP above
Like AID exception using
net ODA from OECD-DAC
log (population)
32
those over 25
Arms imports/total imports, lagged
ARMS-1
U.S. Department of
State, various years
All variables aggregated over time using arithmetic averages.
33
Table A2. Overview of Differences in Regression Samples, Original and Expanded Data
Sets
Regression
Lost in expanded data set
Burnside &
Dollar
Somalia,
Tanzania
Collier & Dehn
““
Collier & Dollar,
Collier &
Hoeffler
None
Hansen & Tarp
Somalia
Dalgaard, Hansen, and Tarp
None
Gained in expanded data set
Burkina Faso, Bulgaria, China, Cyprus,
Hungary, Iran, Jordan, Myanmar, Papua
New Guinea, Poland, Republic of Congo,
Romania, Singapore, South Africa,
Uganda
““
Angola, Burkina Faso, Bulgaria, China,
Cyprus, Czech Republic, Guinea, GuineaBissau, Iran, Jordan, Liberia, Mongolia,
Mozambique, Myanmar, Namibia, Oman,
Papua New Guinea, Poland, Republic of
Congo, Romania, Somalia, Suriname,
Uganda
Angola, Burundi, Benin, Burkina Faso,
Barbados, Bulgaria, Central African
Republic, Chad, China, Cyprus, Hungary,
Iran, Jordan, Mauritania, Mauritius,
Mozambique, Myanmar, Nepal, Papua
New Guinea, Poland, Republic of Congo,
Romania, Rwanda, Singapore, South
Africa, Uganda
Bangladesh, Czech Republic, GuineaBissau, Singapore, South Africa,
Tanzania
Guillaumont &
None
None
Chauvet
Source: Author’s analysis based on sources described in the text.
Study period
Original
Expanded
1970–93
1970–
2001
1974–93
1974–97
1974–97
1974–
2001
1978–93
““
1974–97
““
1970–93
1970–93
34
Appendix 2. Partial scatters of growth against variables of interest
Each figure below is a partial scatter of GDP/capita growth versus a variable of interest in the
context of a tested regression. The figures correspond to OLS and 2SLS results of table 6 in the
main text. Data points are labeled by ISO three-letter codes and the last two digits of the starting
year of the period. Outliers are marked separately, and two partial regression lines are shown,
one for the full sample, one for the sample excluding outliers. Note that the second line in each
graph is not the best fit to the non-outlier data points as plotted. Deleting observations causes the
estimated coefficients to shift and all remaining data points in the partial scatter to move. The
second line is the best fit to the data points in their post-exclusion positions, which are not
shown.
35
Figure A1. B&D regression 5/OLS: Partial scatter of GDP/capita growth versus aid×policy
Original data
GAB74
GDP/capita Growth
-5
0
5
10
CMR78
ZMB86
ZMB90
SYR74
DOM70
EGY82
NGA70
PRY78
EGY74
GAB70BRA70
ECU70
MEX78
SYR90
SYR78
LKA82
NGA90
PAK82
ARG90
MLI86
KEN70
EGY78
KOR86TTO78
BOL82
PAK78
URY86
BRA86
IDN90
SEN82
URY78
KOR82
CMR82
SLV90
THA86
NGA74
DOM86
NGA86
PAK86
LKA90
ARG74
TUR90
IND86
KEN86
ETH86
MAR82
SOM74
MAR74
NIC74
TUN82
CMR74
CHL90KOR90
GAB82
THA90
IDN78
SLE78
IND82
ZWE82
GUY74
PAK90
PRY74
PER78
HTI78
NER78
NGA82
BRA82ARG82
ZAR82
GMB74
GTM70
COL78
IDN86
TZA82
ZWE86
GTM74
GAB90
IDN82
KEN78
CHL86
TGO78
SLV82
MWI82
CRI70
ZAR70
CHL78
LKA78
GMB82
MYS90
URY74PHL78
KOR74
BRA74
ARG86 BOL90
COL70
LKA86
MEX70
COL82
TTO74
SLV74
GHA78
COL90
GHA70
VEN90
TZA86
ARG70
PHL86
COL86
SLE86
PHL74 GHA86
KOR70
PER90
DOM78
URY90
LKA74
MAR78
TUN90
THA82
SEN74
HND78
HND86
PRY70
SLE90
EGY86
SLE70
KEN82
BOL74
SYR70
IND90
DOM74
PER70
DZA74
GUY78
MYS78
PHL70GHA90
GTM78
HND70
ZAR86
PER74
TGO74
HND74
IDN74
MEX82
ZMB70
HND90
DZA70
ARG78
HND82
BRA78
PER86
SLV70
PRY86
MDG86
TGO86
HTI74
PAK70
GHA82
GUY70
NIC82
CRI90
DOM82
MAR86
CRI74
MEX74
GTM86
ZWE90
SLE82
ECU82
GTM90
KOR78
SLV86
BRA90
DOM90
IDN70
COL74
MWI78
MYS82
ZMB78
IND78
PAK74
GAB86
HTI82
ECU90
TTO70
HTI70
TGO82
JAM82
THA78
PER82
ECU78
IND74
LKA70
MWI90
EGY90
MAR70
ECU74
ECU86
MYS70
KEN74
VEN74
MEX90
GMB78
ZMB82
BOL78
GUY86
NIC70
CRI86
MYS86
KEN90
MWI86
CRI82
BOL70
TUN86
BOL86
SEN70
SLE74
CRI78
VEN70
SOM78
URY70
ZAR78
GUY82
ZMB74
VEN86
HTI86
PRY82
MYS74
GTM82
THA70
CHL70
PRY90
GMB70
MDG90
JAM78
CIV78
MDG70
NER74
THA74
NGA78
MEX86
PHL90
VEN82
TTO82
MAR90
SYR86
SEN78
IND70
CHL82
MDG74
CMR86
CHL74
PHL82
GHA74
ZAR74
URY82
SLV78
VEN78
TTO86
JAM74
BWA78
BWA82
BWA86
-10
CMR90
GAB78
NIC78
ETH82
-10
-5
Non-outliers
Outliers
0
Aid*Policy
5
10
Fit to all data, coef = 0.19, t = 2.61, N = 270
Fit excl. outliers, coef = -0.05, t = -0.45, N = 263
10
Expanded sample
GDP/capita Growth
-5
0
5
SYR74
NGA70MMR98
GAB70
CMR78
PRY78
BWA86
EGY74
DOM70
TGO94
BWA78
PNG90
SYR78
IRN90
CMR82
EGY82
CHN94
MEX78
BRA70
CHN90
BOL82
BWA82
KEN70
IRN82 URY86
KOR86
TTO78
SYR90
CHN98
CHL90
BWA74
CYP82
LKA90
BWA98
UGA94
UGA86
TUR90
CYP78
SEN98
CYP86
MAR74
ZWE78
PAK82
UGA90
THA86
PAK78
URY78
CHN86
URY90
ZWE94
BGR98
EGY78
LKA82
MLI86
MMR94
IND98
NGA90
KOR82
MLI98
ARG74
PER94
CHL86
MYS90
IND94
NGA86
CHL94
TTO74
SEN82
THA90
IDN90
CMR74
GHA98
MWI90
IND86
DOM98
MAR82
TUR82
POL94
TUN70
SGP70
BRA94 PER90
GTM78
IND82
PRY74
NGA74
IRN98
HUN98
KOR90
CMR98
ETH94
CYP98
IDN78
UGA98
GMB74
PAK86
TGO78
TUR86
URY94
PAK90
ETH86
MMR90
BRA82
KEN86
LKA78
NGA98
GTM70
URY74
NIC98
MEX98
VEN90
KEN78
SEN94
TUN98
HTI78
COG98
LKA94
DOM86
IND90
MYS94
ARG94
IDN86
NIC74
CHL78
POL98
DZA98
DZA70
BFA94
TUR74
SLE78
TZA98
TUN78
BRA74
TGO74
LKA86
GTM74
MWI82
TTO82
MMR78
ZMB98
MMR82
BFA82
COL90
GHA86
UGA82
GTM90
TUR94
TTO94
TUN90
PER78
ZAR82
CRI70
COL78
GMB82
KOR94
COL94
KOR74
TUN74
COL86
COL70
ZAR94
DOM94
PRY86
MDG98
GAB82
ETH98
TUN82
IDN70
CRI98
SGP86
IDN94
LKA98
EGY98
PHL86
GHA90
CIV94
CYP90
ZAF98
GHA70
ECU70
PRY70
IDN82
ZWE86
EGY90
GAB90
JAM90
ARG78
HND90
MLI94
SLE82
PAK98
DZA74
ARG70
BOL90
GMB78
ZAR86
SGP90
MEX90
MEX70
ROM94
SGP78
SYR70
CRI90
SEN74
BFA98
BWA94
EGY86
BFA90
SYR98
GHA78
MEX94
ZWE82
PHL78
KOR98
BFA74
PHL74
TZA90
SLE70
THA94
SYR94
IDN74
KOR70
HUN86
ZMB70
COL82
PHL70
NER78
PAK94
TGO86
NGA94
GHA94
BRA78
MAR86
ZAR70
BFA86
THA82
IRN94
CYP94
HND78
SLE86
GTM86
GTM94
KEN82
HND86
PAK70
TTO70
TUN94
BRA98
BOL74
DOM74
LKA74
DZA94
HTI74
MEX82
EGY94
SGP94
MAR98
MMR74
ETH90
GTM98
HND70
BOL94
ZAF94
COG90
ZWE90
HND74
MYS78
BWA90
CRI74
MAR78
MEX74
ECU86
HTI98
ZMB78
DOM78
HUN94
THA78
MAR70
PAK74
BOL70
GHA82
BFA78
KEN74
VEN94
NGA82
ECU90
PER70
VEN74
PER74
CHL98
KEN90KEN94
TUR70
MEX86
ECU74
NIC82
ROM98
COL74
CRI86
ECU98
KOR78
SLV98
HND82
SLV70
IND74
TZA94
PER82
MDG86
BOL86
VEN86
ZMB82
VEN70
GAB86
ECU94
HTI70
MYS86
DOM82
CIV98
COL98
JAM82
MLI90
SGP82
IND78
PRY90
BOL98
URY70
TUR98
CRI94
TTO90
MYS70
MWI78 ECU78
JOR98
TUR78
CMR94
MWI86
KEN98
HND98ZAR98
HND94
SLE98
THA70
ZMB74
DOM90
MYS82
NIC70
MAR94
THA74
NER74
PHL94
TGO82
JAM98
PNG82
MDG94
CIV78
ZAR78
SEN70
MYS74
SLE74
TUN86
BGR94 CHL74
GMB70
LKA70
HTI82
BOL78
JOR90
URY98
ZAF90
PER98
JAM78
CHL70
PRY94
VEN98
ARG98
MDG90
COG94
CRI82
CIV86
PNG86
PRY82
JAM94
PNG98
CRI78
ECU82
HTI86
SYR86
MAR90
PHL98
PNG78
MMR70
TGO98
SGP98
PRY98
PNG94
SGP74
GHA74 JOR94
GTM82
MYS98
SLE90
THA98
ETH82
CHL82
IND70
CIV82
VEN82
SEN78
CIV90
HTI94
MDG70
BFA70
PHL90
ZAR90
ZAR74
TTO86
CMR86
VEN78
NGA78
MDG74
HTI90
IDN98PHL82
URY82
IRN86
JAM74 ROM90
GAB78
MMR86
HUN90
CMR90 TGO90 JOR86
SLE94
-10
NIC78
IRN78
-4
-2
Non-outliers
Outliers
0
Aid*Policy
2
4
Fit to all data, coef = -0.04, t = -0.23, N = 436
Fit excl. outliers, coef = -0.27, t = -1.33, N = 426
36
Figure A2. Collier & Dehn regression: Partial scatter of GDP/capita growth vs. Δaid×neg.
shock
GDP/capita Growth
-5
0
5
10
Original data
CMR78
SYR86
GMB90
GAB74
EGY82
PRY78
SYR74
BWA78
EGY74
MEX78
SYR90
NGA86
NGA90
BWA82
ARG90
PAK82
TTO78
SEN82
SLE78
SLV90BOL82
PAK78
LKA82
KOR86
URY78
CMR82
KOR82
ARG74
PAK86
IDN90
GAB82
BWA86
GMB82
URY86
BRA86
NIC74
LKA90
GUY74
KEN86
IDN86
TZA82
ETH86
TUR90
ZWE82
NGA74
TGO78
MAR82
TUN82
PRY74
CHL90
KEN78
GAB90
THA86
IND82
MWI82
CMR74
GHA78
HTI78
IND86
GTM74
ZAR82
PER78
THA90
BRA74
PER74
URY74
KOR90
NGA82
BRA82
KOR74
BOL74
CHL78
GMB74
COL78
PAK90
LKA78
ZWE86
DOM86
NER78
IDN78
SLV82
SYR78
TTO74
GUY90
MAR74
IDN82
GAB86
ARG86
CHL86
TZA86
PHL78
EGY78
EGY86
ARG82
MYS90
LKA74
SLV74
COL82
ZMB74
HND86
COL86
VEN90
GHA90
DOM74
BOL90
PHL74
LKA86
GTM78
DOM78
ZMB86
GHA86
SLE90
COL90
MAR78
GUY78
HND78
URY90
TUN90
HND90
PER90
CRI90
ZMB90GMB78
THA82
GTM90
PHL86
MWI78
TGO86
HND82
ZMB78
HND74
MEX74
NIC82
BRA78
MYS78
ZWE90
SLE82
MEX82
TGO82
CRI74
IND90
GTM86
MWI90
MDG86
PRY86
GHA82
ARG78
ZAR86
DOM82
HTI74
PAK74
COL74
VEN86
ECU82
KOR78
KEN82
TUN86
MYS86
ECU86
SLE86
MYS82
ZMB82
VEN74
IND74
PER86
SOM78
IDN74
DOM90
HTI82
CRI86
MAR86
SLE74
IND78
ECU90
ECU74
THA78
BRA90
ECU78
KEN90
PER82
EGY90
MSEN74
EX90
CIV78
BOL78
CRI82
CHL74
CRI78
NIC86MEX86
JAM82
DZA74
MDG90
MYS74
ZAR74
KEN74
PRY82
THA74
JAM78
GTM82
MWI86
PRY90
HTI86
GUY82
SEN78
ZAR78
NGA78
TTO82
VEN82
CMR86
GHA74
CHL82
GUY86
TTO86
MAR90
TGO74
PHL90
NER74
GMB86
MDG74
PHL82
URY82
SLV78
VEN78
JAM74
MLI86
SOM74
NIC90
SLV86
BOL86
-10
CMR90
NIC78
GAB78
ETH82
-50
0
50
Change in Aid*Negative Shock
Non-outliers
Outliers
100
Fit to all data, coef = 0.04, t = 3.17, N = 234
Fit excl. outliers, coef = -0.06, t = -1.33, N = 224
GAB74
SYR86
-10
-5
GDP/capita Growth
0
5
10
15
Expanded sample
NGA70
JOR78
DOM70
PRY78
ARG90
TGO94
CMR78
GAB70
SYR74
BWA78
BWA86
PNG90
CMR82
EGY82
BRA70
BWA82
CHN94
BWA74
NGA86
CHN90
EGY74
KEN70
MEX78
IRN90
CHL90
IRN82
PAK82
SYR90
CYP82
LKA90
MMR94
GUY94
ZWE78
PAK78
TTO78
ZWE94
UGA90
TUR90
ETH94
CYP78
KOR86
GMB74
MMR82
IND82
MLI86
PER94
IND94
GTM78
KOR82
BRA86 SLE78
UGA86
URY86
CHL94
SEN82
URY78
LKA82
POL94
NGA90
PAK86
TUN70
CHN86
IDN86
BOL82
MWI90
MAR74
MAR82
CYP86
TUR82
THA86
URY90
CHL86
SYR78
SGP70
IDN90
IND86
MMR78
MMR90
CMR74
PRY74
THA90
LKA78
TTO74
GTM90
TGO78
BFA82
MWI82
HTI78
ETH86
BRA94
BFA94
CHL78
IDN78
LKA94
ARG74
BRA90
GTM70
KEN78
EGY78
PAK90
MYS90
SEN94
NGA74
IND90
URY94
DOM94
COL86
ZAR82
KEN86
ZMB90
KOR90
BRA82
NIC90
DZA70
EGY86
DOM86
BFA90
PER90
VEN90
ARG94
GMB82
BRA74
TUR74
COL94
COL90
CRI70
TUN90
COL78
NIC74
UGA82
COL70
HND90
CRI90 BOL74
PRY86
URY74
UGA94
ECU70
MLI94
TUR94
TUN82
TUR86
GHA90
CIV94
GHA86
MYS94
GHA78 BFA86
GTM74
IDN70
PRY70
TUN78
KOR74
ROM94
SGP86
TTO94
ARG86
PER78
IDN94
IDN82
KOR94
ZAR94
LKA86
BWA94
ZMB86
SLE82
GHA70
JAM90
EGY90
MMR74
ZWE82
EGY94
GTM94
TTO82
MEX86
TGO74
JOR82
BOL90
GMB78
GAB82
GTM86
KOR70
PER74
SLE70
NGA94
COL82
PAK94
PER86
TZA90
HND86
TUN94
SGP78
ECU86
MEX70
VEN86
MEX94
GHA94
ZWE86
SYR70
CYP90
NER78
ZAR70
HND78
BFA78
MEX90
PHL86
HTI74
ZAR86
HND70
THA82
DZA94
SYR94
ZMB74
ETH90
LKA74
DOM74
ARG70
ARG82
ZMB70
PAK70
DOM78
BRA78
BFA74
KEN94
BOL94
GAB86
IRN94
HUN94
GAB90
KEN82
ARG78
THA94
HND74
PHL78
BWA90
CYP94
MYS86
ZAF94
MAR86
PHL70
DZA74
TGO86
SGP90
CRI86
GHA82
MLI90
MAR70
ZWE90
BOL70
TUN86
IND74
PAK74
HTI70
PHL74
THA78
SEN74
CRI74
MAR78
IDN74
TUR70
IND78
MYS78
ZMB78
HND82
HUN86
DOM82
NGA82
NIC86CHL74
ECU74
KEN90
SLV70
CMR94
MEX82
BOL86
MWI78
MDG86
ECU94
TTO70
COG90
SGP94
JOR74
MDG94
ECU90
PER70
MEX74
COL74
TUN74
ECU78
CRI94
KOR78
GMB70
NIC82
PRY90
VEN94
ZMB82
KEN74
MAR94
SLE86
CIV78
TGO82
MYS70
HTI82
GMB90
VEN74
PER82
DOM90
SGP82
TUR78
GMB86
PNG82
PHL94
ZMB94
THA70
HND94
PRY94
LKA70
VEN70
JAM82
URY70
MDG90
SEN70
THA74
NIC70
MMR70
SGP74MWI86
SLE74
JAM94
MYS82
ZAR78
ETH82
TTO90
CRI82
PRY82
JOR90
BOL78
MYS74
ECU82
JAM78
TZA94
COG94
CRI78
MAR90
BGR94
PNG94
CHL70
JOR94
ZAR74
CHL82
HTI86
NER74
PNG78
IND70
PNG86
GTM82
NIC94
CIV86
TTO86
BFA70
MDG70
CIV82
HTI94
IRN86
SLE90
SEN78
GHA74
CIV90
VEN82
CMR86
HTI90
MDG74
PHL90
NGA78
VEN78
URY82
PHL82
ZAR90
ROM90
JAM74
MMR86
TGO90
HUN90
CMR90
GAB78
SLE94
JOR86
NIC78
IRN78
-50
Non-outliers
Outliers
0
50
Change in Aid*Negative Shock
100
Fit to all data, coef = 0.03, t = 3.01, N = 391
Fit excl. outliers, coef = -0.10, t = -1.15, N = 372
37
Figure A3. Collier & Dollar regression: Partial scatter of GDP/capita growth vs. aid×policy
20
Original data
GDP/capita Growth
0
10
GAB74
SDN94
SDN74TGO94
BWA82BWA78
PAN90
PRY78 CMR78
NIC94
SYR90
BHS82
BWA86
SYR74
CMR82
EGY82
BHS78
SDN90
TTO78
ARG90
ETH86
EGY74
GUY94
MEX78
GUY90
MLI74
CHL78
SLV90
THA86
ZWE78
JAM86
DOM86
HTI78
MLI86
BWA74
IDN78
KOR82
KOR86
THA90
PER94
IDN90
PAN78
SEN82
SYR78
CHL90
NGA90 CHL86
ETH94
DOM94
NIC74
DOM74
VEN90
DOM78
KEN86
SLE86
SLE78
ECU74
PAK78
LKA82
PAK82
DOM82
PRY86
MAR82
ECU78
GTM90
TGO78
HND86
CRI90
NGA86
ZAR82
PRY74
DZA82
BGD78
GMB82
HTI74
MWI90
EGY78
URY78
SLV82
LKA90
URY86
SDN86
URY90
COL90
IND94
ETH90
IDN94
HUN86
CMR74
MWI82
HUN82
ECU90
IDN74
NER78
MWI94
GAB94
COL86
BOL74
GTM74
ZWE94
MLI82
SLV86
BOL90
IDN82
MYS90
BGD82
BRA86
JAM90
CHL94
CIV94
BRA94
TZA90
IND82
GTM78
HND90
ZMB86
CIV74
THA94
GMB74
NGA74
KOR90
THA82
BGD90
BWA90
PER78
EGY86
TTO82
TZA86
SEN94
TUN90
HUN94
GHA90
CRI86
DOM90
NER86
IDN86
SLV94
NGA94
IND86
MEX90
ZWE86
ZWE82
SLV74
MAR94
KOR94
PHL86
GAB90
MLI94
COL78
BRA82
MAR74
MYS94
GAB82
PRY90
NIC82
BOL86
BRA74
ZAR86
MAR86
PER90
KEN78
GUY74
HTI82
SEN86
ECU82
SDN82
GTM86
URY94
PAK86
GHA86
COL94
URY74
GTM94
TUR94
PAK90
SGP78
ZAR78
PAN82
MWI74
KEN94
HND74
GUY86
EGY90
KEN82
EGY94
BWA94
GHA94
ECU94
HND78
LKA78
COL82
SLE82
MDG82
KOR74
SLE74
NER94
MYS78
THA78
PER86
HND82BOL94
MYS86
SEN74
BGD74
IND90
TZA94
TUN78
SEN90
GMB86
GHA78
DZA74
LKA94
LKA74
ZAF94
BRA78
TTO94
MDG86
NGA82
THA74
PRY94
BGD94
GMB90
GMB78
ARG74
MLI90
TGO86
TUN82
TGO74
CHL82
SGP82
MYS82
TGO82
ARG82
LKA86
ZMB90
KEN90
HTI86
TUN74
GHA82
BRA90
ARG86
BOL82
DZA78
CMR94
ARG94
PHL94
BOL78
ZWE90
NER74
BGD86
TTO90
ETH82
PHL78
MAR78
CRI82
MEX82
GUY78
PAN94
BHS86
ZMB78
TUN94
JAM82
GTM82
SYR82
MEX74
ECU86
MLI78
NER90
IND78
IND74
CRI94
ARG78
PER74
ZMB82
ZAR94
MDG94
PAK74
PAK94
JAM94
PHL74
MDG90
CIV86
CRI74
VEN86
GAB86
MWI86
CIV90
SYR86
MWI78
HTI94
NIC90
MDG78
MAR90
PER82
COL74ZAR74
DZA94
MYS74
MDG74
PRY82
MEX94
JAM78
MEX86
KOR78
ZMB74
HND94
KEN74
HTI90
VEN94
PHL90
PAN74
CRI78
SDN78
HUN90
GUY82 GMB94
BHS90
SGP74
TUN86
CHL74
CIV78
SEN78
CMR86
PAN86
DZA90
DZA86
SLE90
CIV82
NIC86 SLE94
NGA78
NER82
GHA74
TTO86
PHL82
CMR90
URY82
GAB78
JAM74
SLV78
TGO90
NIC78
ZAR90
-10
ZMB94
-15
-10
Non-outliers
Outliers
-5
Aid*Policy
0
5
10
Fit to all data, coef = 0.14, t = 2.15, N = 344
Fit excl. outliers, coef = 0.07, t = 1.06, N = 341
20
Expanded sample
LBR98
GAB74
LBR94
GDP/capita Growth
-20
0
CYP74
COG78
OMN82
COG82
SDN74
CHN82
SYR90BWA86
SDN86
PAN90
ARG90
MOZ98
PRY78
MOZ86
AGO94
BHS82
BWA82
CHN90
CMR78
SYR74
PNG90
BWA78GUY90
SDN98
TTO78
BHS78
TGO94
IRN90
MEX78
SDN90
SLV90
CMR82
EGY82
CHN94
EGY74
CHL90
SDN94
THA86
THA90
MMR90
CYP86
SOM82
AGO98
ETH86
CHL78
IDN90
MMR94
GUY94
TUR90
MOZ94
BGR98
DOM86
DOM98
CHN98
MLI74
CHN78
IDN78
JAM86
ZWE78
KOR86
HTI78
VEN90
BFA82
UGA94
NGA90
SYR78
MLI98
SUR86
ROM82
MYS90
DOM94
LKA90
CHL86
CHN86
GTM90
POL94
NIC98
AGO86
KOR82
CYP82
TTO98
PER94
PAN78
UGA90
MLI86
BWA74
CRI90
PAK78
CYP78
SOM86
SLE78
UGA86
CMR98
BWA98
DOM74
LKA82
JAM90
MWI90
NIC74
SEN82
SUR90
URY90
PAK82
PRY86
CHL94
URY86
COL90
KOR90
DOM78
SEN98
GAB94
URY78
GMB98
ETH94
MMR82
TGO78
ECU74
ZWE94
ECU78
COG98
EGY78
HTI74
NGA86
GHA98
KEN86
PRY74
MDG98
BGD90
BFA86
HND86
ZAR82
MAR82
TZA98
MMR78
IDN74
MEX90
UGA98
DOM90
ETH98
GTM74
GIN98
CYP90
GAB90
DOM82
GNB82
CMR74
MEX98
BOL90
GIN90
TUN90
GTM78
GIN94
PAK90
GIN86
BRA86
TTO82
NGA74
COL86
TUR82
URY94
TUR74
GHA90
HND90
SLV82
SLV86
GNB90
DZA82
IDN82
ECU90
IDN94
GUY74
PHL86
CIV74
IDN86
SLE82
NER98
BRA94
HUN98
NGA98
CRI98
TTO94
BFA98
NER78
GAB82
GNB94
MAR74
SGP78
ETH90
ARG94
COG90
BOL74
IND98
BGD98
MOZ90
PRY90
THA82
PAN98
BWA90
BFA90
GMB82
IND94
PER78
EGY86
MYS94
GUY86
PAK86
GMB74
ZMB86
SEN94
URY74
BGD82
BGD78
CIV94
MWI94
IND82
SLE86
GTM86
BFA74
IND90
OMN90
HUN86
EGY90
CRI86
UGA82
EGY98
SLV74
MWI82
ECU94
POL98
SLV94
LKA78
TZA90
IND86
ZAR86
TUN98
ROM94
GTM98
BGD74
BFA94
DZA98
GTM94
KOR94
HUN94
BFA78
ECU98
BRA74
TUR86
SLE74
BRA90
NGA94
COL78
MAR86
ZAR78
ZWE86
BRA82
TTO90
MYS86
LKA94
KOR98
THA78
MYS78
TUR94
GHA86
TGO74
NAM94
PER86
LKA86
TZA86
MLI94
KEN78
HND74
THA94
HND78
ROM86
OMN78
SEN90
SEN86
KOR74
EGY94
CRI94
ZWE82
PAN82
COG74
ARG74
NER86
JOR90
COL94
HUN82
MMR74
PER90
HTI98
GNB78
BWA94
CMR94
SEN74
ROM98
ZMB98
DZA74
KEN82
THA74
HTI82
LKA74
NIC82
BHS86
HND82
MLI90
BRA78
PRY94
MEX94
TZA94
SUR78
ZAF94
MAR98
SLV98
BGD94
BOL94
GMB90
NGA82
GHA78
TUN78
GHA94
MWI74
KEN90
JAM98
NAM98
VEN94
COL82
KEN94
KEN98
SLE98
PAN94
SGP82
ZWE90
BRA98
PRY98
MDG82
ARG78
MYS82
POL90
GNB86
CZE94
MDG86
ARG86
LKA98
GUY98
ZAF98
ECU86
ECU82
SDN82
NER94
CZE98
ZMB90
DZA78
PHL94
BOL98
GUY78
GMB78
TUN94
CHL82
PHL78
MEX82
BGD86
TUN74
GAB86
BOL82
MNG98
TUN82
JAM94
BOL86
HND98
HTI86
ZMB78
PAK74
MNG94
MEX74
GTM82
CIV90
ARG82
PAK98
MWI98
SYR82
TGO86
GMB86
VEN86
GAB98
BOL78
NER90
SYR86
PER74
CIV98
VEN98
NIC94
MAR90
PHL74
GHA82
MAR94
IND78
IND74
MDG90
PHL90
NER74
IRN94
CIV86
CHL98
CRI82
JAM82
LBR74
BHS90
MDG94
JOR82
MEX86
JOR98
HND94
SUR82
NIC90
KOR78
CRI74
PER82
MLI78
TGO82
ETH82
DZA94
PER98
COG86
ZMB82
PNG82
MYS74
ZAR74
PAK94
MDG78
JAM78
COL74
PRY82
ZMB74
MDG74
HTI90
PHL98
ZAR98
SLE90
SOM74
GUY82
LBR82
MAR78
LBR78
JOR94
COG94
MLI82
ROM78
SGP74
MWI86
DZA90
OMN86
TGO98
COL98
PAN74
GMB94
PNG86
KEN74
SDN78
URY98
MWI78
TUR78
CRI78
PNG78
THA98
CIV78
ZAR94
PNG98
BGR94
PAN86
NIC86
CHL74
TUR98
MYS98
TUN86
SOM90
ARG98
SEN78
DZA86
IDN98
ZWE98
TTO86
HTI94
CMR86
PNG94
NGA78
CIV82
CMR90
MMR86
GAB78
BGR90
GHA74
ROM90
GNB98
PHL82
HUN90
ZMB94
CZE90
NER82
URY82
JAM74
SOM78
TGO90
MOZ82
LBR86
SLV78
PNG74
AGO90
NIC78
ZAR90
SLE94
MNG90
JOR86
BHS74
JOR74
JOR78
-40
LBR90
-60
-40
Non-outliers
Outliers
-20
Aid*Policy
0
20
40
Fit to all data, coef = -0.01, t = -0.19, N = 520
Fit excl. outliers, coef = -0.04, t = -0.81, N = 508
38
Figure A4. Collier & Hoeffler regression: Partial scatter of GDP/capita growth vs. Postconflict 1×aid ×policy
20
Original data
GDP/capita Growth
0
10
GAB74
SDN94
SDN74TGO94
BWA82BWA78
PAN90
PRY78 CMR78
NIC94
SYR90
BHS82
SYR74 BWA86
CMR82
EGY82
BHS78
SDN90
TTO78
ARG90
ETH86
EGY74
GUY94
MEX78
GUY90
MLI74
CHL78
SLV90
THA86
ZWE78
JAM86
DOM86
HTI78
MLI86
BWA74
IDN78
KOR82
KOR86
THA90
PER94
IDN90
PAN78
SEN82
SYR78
CHL90
NGA90 CHL86
ETH94
DOM94
NIC74
DOM74
VEN90
DOM78
KEN86
SLE86
SLE78
ECU74
PAK78
LKA82
PAK82
DOM82
PRY86
MAR82
ECU78
GTM90
TGO78
HND86
CRI90
NGA86
ZAR82
PRY74
DZA82
BGD78
GMB82
HTI74
MWI90
EGY78
URY78
SLV82
LKA90BOL90
URY86
SDN86
URY90
COL90
IND94
ETH90
IDN94
HUN86
CMR74
MWI82
HUN82
ECU90
IDN74
NER78
MWI94
GAB94
COL86
BOL74
GTM74
ZWE94
MLI82
SLV86
IDN82
MYS90
BGD82
BRA86
JAM90
CHL94
CIV94
BRA94
TZA90
IND82
GTM78
HND90
ZMB86
CIV74
THA94
GMB74
NGA74
KOR90
THA82
BGD90
BWA90
PER78
EGY86
TTO82
TZA86
SEN94
TUN90
HUN94
GHA90
CRI86
DOM90
NER86
IDN86
SLV94
NGA94
IND86
MEX90
ZWE86
ZWE82
SLV74
MAR94
KOR94
PHL86
GAB90
MLI94
COL78
BRA82
MAR74
MYS94
GAB82
PRY90
NIC82
BOL86
BRA74
ZAR86
MAR86
PER90
KEN78
GUY74
HTI82
SEN86
ECU82
SDN82
GTM86
URY94
PAK86
GHA86
COL94
URY74
GTM94
TUR94
PAK90
SGP78
ZAR78
PAN82
MWI74
KEN94
HND74
GUY86
EGY90
KEN82
EGY94
BWA94
GHA94
ECU94
HND78
LKA78
COL82
SLE82
MDG82
KOR74
SLE74
NER94
MYS78
THA78
PER86
HND82BOL94
MYS86
SEN74
BGD74
IND90
TZA94
TUN78
SEN90
GMB86
GHA78
DZA74
LKA94
LKA74
ZAF94
BRA78
TTO94
MDG86
NGA82
THA74
PRY94
BGD94
GMB90
GMB78
ARG74
MLI90
TGO86
TUN82
TGO74
CHL82
SGP82
MYS82
TGO82
ARG82
LKA86
ZMB90
KEN90
HTI86
TUN74
GHA82
BRA90
ARG86
BOL82
DZA78
CMR94
ARG94
PHL94
BOL78
ZWE90
NER74
BGD86
TTO90
ETH82
PHL78
MAR78
CRI82
MEX82
GUY78
PAN94
BHS86
ZMB78
TUN94
JAM82
GTM82
SYR82
MEX74
ECU86
MLI78
NER90
IND78
IND74
CRI94
ARG78
PER74
ZMB82
ZAR94
MDG94
PAK74
PAK94
JAM94
PHL74
MDG90
CIV86
CRI74
VEN86
GAB86
MWI86
CIV90
SYR86
MWI78
HTI94
NIC90
MDG78
MAR90
PER82
COL74ZAR74
DZA94
MYS74
MDG74
PRY82
MEX94
JAM78
MEX86
KOR78
ZMB74
HND94
KEN74
HTI90
VEN94
PHL90
PAN74
CRI78
SDN78
HUN90
GUY82 GMB94
BHS90
SGP74
TUN86
CHL74
CIV78
SEN78
CMR86
PAN86
DZA90
DZA86
SLE90
CIV82
NIC86 SLE94
NGA78
NER82
GHA74
TTO86
PHL82
CMR90
URY82
GAB78
JAM74
SLV78
TGO90
NIC78
ZAR90
-10
ZMB94
-15
-10
Non-outliers
Outliers
-5
Aid*Policy
0
5
10
Fit to all data, coef = 0.14, t = 2.15, N = 344
Fit excl. outliers, coef = 0.07, t = 1.06, N = 341
20
Expanded sample
GDP/capita Growth
-20
0
LBR98
GAB74
LBR94
CYP74
COG78
OMN82
COG82
SDN74
BWA86
SYR90
CHN82
SDN86
PAN90
ARG90
MOZ98
MOZ86
BWA82
PRY78
BHS82
AGO94
PNG90
CMR78
CHN90
SYR74
BWA78
TGO94
TTO78
SDN98
BHS78
GUY90
MEX78
JOR78 GUY94
SLV90
IRN90
SDN90
EGY82
EGY74
CHN94
CMR82
SDN94
CHL90
THA90
THA86
SOM82
CYP86
MMR90
AGO98
MMR94
BGR98
IDN90
CHL78
ETH86
MLI74
TUR90
MOZ94UGA90
DOM98
CHN98
DOM86
JAM86
IDN78
CHN78
HTI78
SYR78
UGA94
BFA82
ZWE78
KOR86
VEN90
NIC98
MLI98
SUR86
LKA90
DOM94
NGA90
BWA74
POL94
GTM90
ROM82
MYS90
MLI86
TTO98
CHN86
PAN78
CYP82
KOR82
CHL86
AGO86
CYP78
SOM86
PER94
CRI90
PAK78
MWI90
UGA86
DOM74
JAM90
LKA82
CMR98
BWA98
SLE78
NIC74
SEN82
SUR90
PAK82
URY90
GAB94
SEN98
DOM78
PRY86
COL90
TGO78
CHL94
URY86
KOR90
MMR82
URY78
GMB98
ECU74
ETH94
ECU78
COG98
EGY78
HTI74
HND86
ZWE94
BFA86
MDG98
GNB82
PRY74
IDN74
MAR82
MMR78
TZA98
BGD90
GHA98
KEN86
NGA86
CYP90
ZAR82
GIN90
BOL90
GIN98
UGA98
DOM82
GAB90
DOM90
MEX90
GTM74
GIN94
MEX98
GNB90
CMR74
GTM78
ETH98
GIN86
TUN90
SLV86
TTO82
PAK90
GNB94
HND90
SLV82
TUR74
TUR82
BRA86
GUY74
GHA90
URY94
PHL86
IDN82
ECU90
NGA74
COL86
IDN94
HUN98
DZA82
GAB82
IDN86
GMB82
MOZ90
NER78
TTO94
BFA98
BRA94
CIV74
NER98
CRI98
SLE82
BFA90
BOL74
ARG94
ETH90
BGD98
IND98
COG90
NGA98
MAR74
BWA90
PRY90
THA82
SGP78
SEN94
GMB74
ZMB86
EGY86
GUY86
MYS94
PER78
PAN98
MWI94
CIV94
IND94
PAK86
BGD78
BGD82
BFA74
URY74
SLE86
GTM86
EGY90
TZA90
LKA78
CRI86
POL98
OMN90
HUN86
IND82
BFA94
SLV74
MWI82
ECU94
UGA82
IND90
SLV94
BGD74
EGY98
ROM94
GTM98
BFA78
GTM94
TUN98
HUN94
ECU98
KOR94
ZAR86
IND86
DZA98
KOR98
TUR86
MLI94
SEN86
TZA86
LKA94
BRA74
LKA86
PER86
SLE74
TTO90
MYS86
ZWE86
BRA82
BRA90
TGO74
GHA86
SEN90
NAM94
TUR94
MAR86
THA78
COL78
ROM86
ZAR78
NGA94
NER86
HND74
OMN78
THA94
MYS78
JOR90
EGY94
HTI98
HND78
ARG74
COG74
KEN78
ZWE82
ZMB98
KOR74
MMR74
ROM98
CRI94
PAN82
HUN82
SEN74
CMR94
BWA94
HTI82
COL94
PER90
LKA74
SUR78
MLI90
THA74
NIC82
GMB90
HND82
DZA74
KEN82
BHS86
BOL94
TZA94
MEX94
GNB86
PRY94
BGD94
VEN94
BRA78
MWI74
GHA94
MAR98
PRY98
SLV98
KEN90
NAM98
TUN78
BRA98
JAM98
GHA78
SGP82
ZAF94
NGA82
ZMB90
GUY98
PAN94
ZWE90
ARG78
KEN94
POL90
SLE98
MYS82
COL82
KEN98
GMB78
CZE94
CZE98
LKA98
MDG86
ARG86
MDG82
NER94
GUY78
BOL98
SDN82
ECU82
ECU86
PHL94
ZAF98
MNG98
DZA78
BOL82
TUN74
HND98
BGD86
GAB86
MEX82
GMB86
MNG94
BOL86
JAM94
TUN94
PHL78
CHL82
TGO86
ZMB78
HTI86
TUN82
PAK74
MWI98
SYR82
PAK98
CIV90
GTM82
ARG82
BOL78
NER90
MEX74
SYR86
VEN98
GAB98
JOR82
PER74
CIV98
VEN86
NER74
PHL74
PHL90
JAM82
CRI82
MDG90
IND74
MAR90
NIC90
MAR94
CHL98
IND78
GHA82
CIV86
HND94
MDG94
LBR74
IRN94
SUR82
PNG82
JOR98
BHS90
MLI78
TGO82
PER82
MEX86
ZMB82
DZA94
CRI74
KOR78
PER98
JAM78
MYS74
ETH82
COG86
PAK94
ZAR74
ZMB74
PRY82
MDG78
LBR82
COL74
HTI90
PHL98
MLI82
SOM74
COG94
GUY82
MDG74
SLE90
JOR94
MAR78
LBR78
ZAR98
MWI86
PNG86
GMB94
SGP74
PAN74
COL98
ROM78
SDN78
URY98
OMN86
DZA90
TGO98
MWI78
KEN74
TUR78
PNG78
CRI78
THA98
PNG98
CIV78
BGR94
NIC86
ZAR94
TUR98
MYS98
PAN86
SEN78
CHL74
TUN86
IDN98
ARG98
SOM90
HTI94
PNG94
DZA86
TTO86
ZWE98
CMR86
NGA78
CIV82
BGR90
CMR90
GNB98
GAB78
MMR86
ZMB94
ROM90
GHA74
PHL82
HUN90
NER82
CZE90
URY82
JAM74
TGO90
SOM78
MOZ82
LBR86
SLV78
PNG74
AGO90
NIC78
SLE94
MNG90
ZAR90
JOR86
BHS74
JOR74
NIC94
GNB78
-40
LBR90
-10
0
Non-outliers
Outliers
10
20
Post-conflict 1*Aid*Policy
30
40
Fit to all data, coef = 0.08, t = 1.91, N = 520
Fit excl. outliers, coef = -0.06, t = -0.19, N = 494
39
Figure A5. Guillaumont and Chauvet Regression 1.2: Partial Scatter of GDP/Capita
Growth versus Aid×Environment
4
Original data
GDP/capita Growth
-2
0
2
GMB82
GMB70
IND82
THA82
TTO70
DOM70
KOR82
PRY70
PAK82
ECU70
GTM82
GHA82
IDN70
MWI82
BOL82
SLV82
SYR82
EGY82
MYS82
DOM82 IDN82
HND82KOR70COG82
BRA70RWA70
KEN82
VEN82
COL82 JAM82
PAK70
PRY82 CYP82
HND70
URY82
MYS70
CRI82
MUS70
SLE70
CHL82
GUY82
MUS82
SLE82
MEX82
ARG82
PHL82
GUY70
PHL70
TGO82
VEN70 RWA82
BRA82
URY70
COL70BOL70
ECU82 TTO82
THA70
CHL70 LKA82
MMR82SLV70
GHA70
NIC82
-4
SEN82
CMR82
JOR82
ZAR82
-15
-10
Non-outliers
Outliers
-5
0
5
Aid*Environment, instrumented values
10
Fit to all data, coef = -0.15, t = -1.79, N = 68
Fit excl. outliers, coef = -0.11, t = -1.96, N = 67
40