Faculty of Economics and Business Bachelor Thesis
Determinants of the Netherlands’ Aid Allocation Decisions
July 2014
Leonard Treuren
10087621
BSc Economics and Business
Specialization: Economics
Abstract
Given the recent public debate regarding the development aid budget in the Netherlands it is worthwhile to take a step back and describe the actual aid allocation process in detail to determine if actual decisions are in line with stated goals. The official statement regarding development aid stresses the need of the recipient nation and the recipient nation’s ability to generate growth, so as to potentially become a trade partner in the long run. However, the literature stresses more directly selfish motives as driving aid allocation. This seeming contradiction is investigated. The dataset spans the period 1980-2010 and focuses exclusively on the Netherlands, making it stand out in aid allocation literature. The methodology is that of a panel data regression employing a two-stage model with a Heckman correction. The results point towards the Netherlands being driven almost exclusively by altruistic motives in its aid allocation decisions. To further its stated goal of transforming need relationships into trade relationships the Netherlands might do well by placing more weight on the ability of recipient nations to generate growth following development assistance.
Table of Contents
I Introduction 2 II Literature Review 4 III Methodology 7 A. Research Outline 7 B. Data 8 C. Method 12IV Analysis and Results 16
A. Descriptives 16
B. Regression Results and Interpretation 18
VI Conclusion 25
VII Bibliography 27
VIII Appendices 30
A. Variable Specification 30
B. Countries and Codes 31
I. Introduction
Successful economic recovery in Europe following World War II has often been credited to the Marshall Plan. The United States gifted thirteen billion dollars’ worth of development aid, over one hundred billion in current dollars, to Europe in order to foster rebuilding of the economy and to prevent Soviet communism from getting a grip on continental Europe (Ketz, 2002, p. 95). After the plan’s success, development aid became more and more a central feature of poverty alleviation in developing countries. Last year alone, members of the leading organization for development aid, the OECD’s Development Assistance Committee, distributed almost 135 billion total official development aid (“Aid to Poor..”, 2013). Unsurprisingly, the need of developing countries has always been explicitly at the forefront of aid debates. However, by some countries the self-interested nature of aid allocation has not been denied as governments often justified aid expenses on political grounds. Cold War concerns especially led the United States, and in its wake much of the developed world, to embrace the strategic use of foreign aid to forge political alliances. By the end of the Cold War development aid had become an institutionalized aspect of relations between rich and poor countries (McKinley, 1979, p.236).
The Netherlands has been one of the leading donors of development aid throughout the century. As of 2013, the Netherlands remain in the top ten donors based on total development aid as well as development aid as a percentage of GDP (“Aid to Developing..”, 2014). However, since the global financial crisis of 2008 the call for budget cuts, fuelled by increasing public debt, has been increasingly heard worldwide and the Netherlands is no exception. The ministry of foreign affairs, the ministry dealing with development assistance, has set as its goal the reduction of its budget by 25 percent by 2018 (“Ministry of..”, 2013). Overall sentiments in politics have not been favourable towards development aid with as a prime example the VVD, the liberal party, calling for drastic budget cuts in this area and an outright stop to money allocated to multilateral aid organizations (Dool, 2012). The increasing pressure on development aid in the Netherlands makes it interesting to
investigate why exactly the Netherlands gives aid in the first place.
A quick look at the official statements of the ministry of foreign affairs shows a much more altruistic vision on development aid than mentioned above. The Dutch foreign aid policy focuses on four points: security and rule of law, water management, food supply, and sexual and reproductive health and rights. However, the main framework in which the Dutch government makes its aid decision is based upon a combination of development aid and trade (Rijksoverheid, 2014). The goal is to transfer relationships based on need into trade relationships whereby both countries can benefit economically and socially. Initially, aid is allocated to countries most in need. Once aid has proven to be effective the country can become a trading partner and an “equal” party in the exchange. This seems to imply allocation of aid based on need and possible on potential effectiveness of aid, merit, given the ultimate goal of changing the need relationship into a trade relationship. The relevance of trade is slightly ambiguous seeing as the official statement seems to call for trading with nations that have received aid rather than offering developmental assistance to trade partners. As we will see, aid allocation literature usually finds other factors of self-interest to play a major role
This study attempts to investigate the motives for the Netherlands’ aid allocation regardless of official governmental statements. For the current debate regarding aid budget it is useful to first take a closer look at the exact reasons for giving aid. In doing so, this study hopes to clarify current aid allocation decisions made by the Netherlands and provide insights that may prove useful in future debates regarding broader official development aid issues.
II. Literature review
From the 1960s onwards, interest arose regarding the quantitative investigation of the allocation of development aid. While some studies sought to subjectively evaluate this issue, the focus here lies on descriptive studies of aid allocation.
During Cold War-times, the theoretical underpinnings of aid allocation studies could be traced to international relations theory (McGillivray and White, 1993, p.2). Countries often embraced political motives for aid allocation in their policy statements and it was widely accepted that these motives, paired with certain humanitarian considerations, provided a solid a priori basis on which to explain allocation of development aid.
Most early studies employed multiple regression analysis using cross-country data and OLS estimation (i.e. Levitt, 1968). After the initial use of mainly hybrid models with broad specifications the 70s saw the emergence of the recipient need/donor interest models as the dominant paradigm in aid allocation literature (McGillavray and White, 1993, p. 10). This practice involved estimating two separate models, one containing solely recipient need variables and one containing solely donor interest variables. The two models would sub sequentially be tested against one another, with the donor interest model usually providing a better explanation of aid allocation. Influential examples are McKinley and Little (1979) and Maizels and Nissanke (1984). The prolonged dominance of this approach can be considered surprising as earlier use of hybrid models had shown that there was no reason a
priori not to expect that both donor interest and recipient need play a role in aid allocation.
Thus, excluding one of those categories in the model leads to specification bias (McGillivray, 2003, p.4). This equally holds for other approaches investigating limited areas of determinants of aid allocation, without sufficiently controlling for other effects. Examples of such models are the so-called “bias models” which investigate the bias in aid allocation per capita towards small countries and developmental models which focus solely on recipient need variables.
However, such approaches never achieved the level of acclaim received by the RN/DI approach. A separate strain of models, the incremental models, investigate the tendency of bureaucratic decision-makers to base current decisions on last year’s decisions (for instance Gang and Khan, 1990). This approach relies on time series data and is based on slightly different theoretical underpinnings. Although unmistakeably valuable, such studies are relatively few in numbers and shall not be considered here any further seeing as they depart from the approach of this study.
From early 90s the main topics of discussion became econometrical in nature. After Trumbull and Wall (1994), panel data became the dominant type of data and this provided several problems. On the one hand, fixed effects were shown to be relatively important in descriptive studies of aid allocation. This is in part due to the limited availability of data on institutional variables for developing countries. On the other hand, the limited nature of the dependent variable was recognized. Aid allocated cannot be negative rendering linear regression and OLS estimation unsuitable for investigating aid allocation. Separately, this does not prove to be problematic. However, finding a solution enabling fixed effects and dealing with the limited nature of the dependent variable, the so-called “incidental parameter problem”, has proven to be far from straightforward (Berthelemy, 2006b, p.84). No consensus has been reached in the literature as to which modelling approach is to be preferred. Most models have split the decision-making process up into two stages, the selection stage indicating whether a country receives aid, and the allocation stage, indicating how much aid the recipients selected in the first stage will receive. Such models will be discussed in detail later on.
An interesting feature of modern aid allocation literature is the tendency to exclude any theoretical models and build a purely empirical model solely on introspection and empirical findings of others. Prominent examples include Alesina and Dollar (2000) and most of the work of Berthelemy. Notable exceptions include Younas (2008) and Trumbull and Wall (1994), who model aid allocation as a maximization problem with a predetermined aid budget constraint. This seeming lack of theoretical foundations makes certain features of
the econometrical approach appear rather random. The choice of variables is usually loosely based on the earlier RN/DI variables in a hybrid model with additional variables termed as recipient merit (see for instance Hoeffler and Outram, 2011), since Burnside and Dollar (2000) showed good policy to be a determinant of aid effectiveness. Supposedly, the choices of RN/DI variables were once motivated by the political economy theories of international relations but a clear theoretical structure is seldom specified leading to choice of variables and introduction of new variables especially to seem rather arbitrary.
Given the numerous debates it is perhaps surprising that the results of aid allocation studies have been mostly in line with one-another. Earlier studies were somewhat harsher in shooting down recipients need as a determinant of aid allocation (McKinlay and Little, 1979, p. 243) and mostly found political interest to be the main determinant of aid allocation. With the rise of hybrid models as the dominant methodology the findings shifted slightly to also include recipient need and merit as important determinants (Alesina and Dollar, 2000). This has persisted to date with only slight differences in factors identified as important determinants, despite the methodological differences among authors. Another finding that persists in the literature is the so-called “small country effect”. Countries with smaller populations are found to receive higher per capita aid. This is sometimes explained as donors’ preference for countries in which impact of aid is plausibly higher (Alesina and
Dollar, 2000, p.38).Finally, from early publications onwards a difference in aid allocation has
been observed between bilateral and multilateral aid. Multilateral aid has been found to be allocated to a much higher degree based on recipient need than bilateral aid. Often an outcome of voting by members, multilateral aid allocation is thought to reflect individual countries’ behaviour les accurately than bilateral aid (Maizels and Nissanke, 1984, p.891). Given that most studies sought to explain allocative behaviour of individual countries they focused on bilateral aid allocation. This study will follow that approach.
III. Methodology
A. Research Outline
Development aid is defined as Official Development Aid (ODA). It includes loan components, grant components, and technical assistance based on concessional terms (Younas, 2008, p. 661). Aid from a country can usually be divided into aid from NGOs, multilateral governmental aid, and bilateral governmental aid. Roughly 35 percent of Dutch governmental aid goes to multilateral institutions such as the UN and the EU (Rijksoverheid, 2014). In such organizations aid allocation is usually determined by a process involving many nations. Because this paper seeks to address the allocation decisions of the Netherlands alone, solely bilateral ODA will be investigated. The OECD’s Development Assistance Panel (DAC) classification for developing countries is followed. In order for the estimation results to be sufficiently strong the period 1981 through 2010 will be investigated.
Seeing as the target of the Netherlands’ aid programme is to change relationships based on need into relationships based on trade, one would expect aid to target poor countries with potential to grow. However, based on the literature one would expect the Netherlands to equally embrace more selfish motives. This paper seeks to analyse which of the three factors, recipient need, recipient merit, and donor characteristics, affect aid allocation and to what degree they do so.
The method follows the tradition of empirical models in modelling aid allocation, no explicit theoretical structure is presumed. A comprehensive model is amassed in order to investigate whether allocation decisions were purely based on recipient need and merit, or if other factors also played a role. Loosely following a tradition in the literature (see for instance Hoeffler and Outram, 2011) the explanatory variables are grouped into recipient need, recipient merit, and donor characteristics.
B. Data
The dependent variable
In the literature there has been widespread debate regarding whether a per capita measure or an absolute measure with population as a regressor is more suitable as dependent variable, and whether aid commitments or aid disbursements should be used. It should be kept in mind that the ultimate aim of this study is to capture the decision-making process the Netherlands faces as closely as possible. Therefore, this study uses total bilateral ODA commitment as the aid variable. Commitments are chosen over disbursements because commitments reflect more closely the actual intentions of the donors whereas disbursements also reflect the willingness and ability of the recipient to receive aid (McGillivray and White, 1993, p.34). The choice of total ODA commitments relies on the fact that when modelling aid allocation, one should keep in mind the actual decision-making process of the donor. It seems hard to argue that a donor would allocate aid on a per capita basis. Firstly, aid is allocated from a fixed budget in the Netherlands, making it intuitively logical to allocate absolute values. Secondly, allocating per capita would imply decisions regarding differences of less than a cent per capita. It seems unlikely that donors behave in this way when determining aid allocation (McGillivray and White, 1993, pp.34-36). Bilateral aid is examined because it is more directly controlled than multilateral aid, which is often allocated based on consensus between the members. Incidentally, as there is a time lag between aid commitment and aid disbursement the risk of simultaneous causality is reduced (Berthelemy, 2006b, p.82) Aid data were obtained from the OECD’s DAC database (OECD, 2014).
Recipient need
Recipient need is covered by two variables: GDP per capita for the recipient country, and amount of ODA commitments from other donors. GDP per capita is widely accepted and used as a, somewhat crude, indicator of recipient need (for instance Clist, 2011). While it is often seen as a far from perfect indicator of need, need indices are usually heavily correlated with GDP per capita, as McGillivray (1991, p.1461) points out. Furthermore, GDP
per capita is likely to be a variable used in the actual allocation process. Regarding ODA commitments from other donors there is some debate in the literature. Berthelemy (2006a, p.168) argues that one could interpret increased total ODA as a sign of good governance, resulting in the variable being a proxy for recipient merit rather than recipient need. However, following Hoeffler and Outram (2011, p.241) the variable is included as a proxy for recipient need. The precise category of the variable is of little consequence for this study as the point of interest is mainly on recipient need and merit vs. donor self-interest. GDP data are taken from the World Bank (World Bank, 2014). Data on ODA commitments were obtained from the OECD’s DAC database (OECD, 2014).
Recipient merit
Recipients merit is covered by GDP growth per capita, the Political Terror Scale (PTS) and the Freedom House political rights and civil liberties indices. Generally, recipients merit is taken to mean the total of institutions possibly affecting the effectiveness of aid. Since Burnside and Dollar (2000) good policy, policy leading to growth, is believed to positively affect aid effectiveness. GDP growth per capita suffers from certain interpretational problems as it is not exactly clear whether growth indicates need or merit. A donor could see high growth as an indication of sound institutions and governance, but might also see low growth as a sign of need. Again, it is largely irrelevant in which category this variable is placed but following Hoeffler and Outram (2011, p. 241) it is taken as a proxy for merit. The Political Terror Scale and the Freedom House democracy index are taken as measures of civil and political liberties following Alesina and Dollar (2000). Civil liberties and political rights data were obtained from the Freedom in the World report (Freedom House, 2014). The Political Terror Scale was obtained from Gibney, Cornett, Wood, & Haschke (2014). GDP data were obtained from the World Bank (World Bank, 2014). In order to deal with the categorical nature of the Political Terror Scale and the civil liberties and political rights indices two dummies were constructed measuring whether an observation lies above or below the mean of the Political Terror Scale and the mean of the average of the two Freedom House indices.
Donor characteristics
Donor characteristics comprise factors not covering the need for or ability to deal with aid by recipient countries. They are covered by total trade flows between donor and recipients as a
percentage of donor GDP, a dummy indicating if the recipient was a colony of the donor country since 1900, and a dummy indicating donor involvement in major conflicts in recipient territory. The trade variable is a measure of commercial linkages between donor and recipient. A self-interested donor is thought to allocate more aid to countries it has strong commercial ties to as this might positively influence trade (Berthelemy and Trichit, 2004). If the Netherlands follows its own aid statement, this might be especially relevant. In addition, donors might allocate aid favourably to former colonies based on political ties and commercial privileges which usually arise during such relationships. These relationships are commonly thought to run deeper than simple trade relationships. Therefore, following Alesina and Dollar (2000, p.36) colonial status is included as an indicator of self-interest. Wars possibly create special relationships between donor and recipients due to, for instance, reliance. A dummy variable is included indicating whether the Netherlands was involved in a conflict as classified by Correlates of War (2014) in recipient countries ending no earlier than 1900. Trade flows data were obtained from Barbieri and Keshk (2012). Colony data were obtained from Correlates of War 2 Project (2014), and conflict involvement data were obtained from Sarkees and Wayman (2012).
Controls
Population and total size of the aid budget of the Netherlands are included as a control variable. The inclusion of population accounts for its effect on aid allocation and is necessary seeing as the dependent variable is specified in absolute terms. There is much ambiguity in the literature regarding the exact classification of the population variable and the interpretation of its coefficient. It could be included as a measure of recipient need as larger developing countries need more aid, ceteris paribus. It could also be included as a measure of donor interest as donors might be more willing to aid potentially powerful nations politically speaking (Maizels and Nissanke, 1984, p. 881). Due to the unclear nature of
population combined with its significance as an explanatory variable usually reported, population is included as a control variable following Berthelemy (2006b). The total size of the aid budget, again following Berthelemy (2006b), is included to control for scale effects. This has the added benefit of making estimation with this specification identical to specifications using aid per capita or aid as a fraction of total aid as dependent variable as long as logarithms are used and the necesary controls are included.
Lagging variables
Discussions regarding simultaneity have prevailed in the literature. Especially recipient variables are thought to potentially suffer from endogeneity (see for instance Younas, 2008 ). Instrumental variable regression might prove to be a solution but due to the limited access to data in many developing countries finding suitable instruments for all recipients is complicated (Younas, 2008, p.665). The usual solution is lagging certain variables, for instance GDP per capita (Hoeffler and Outram, 2011, p.241). Again, the focus of modelling aid allocation should be on the actual decisions policy-makers face. Because decisions are not made instantaneously, it is safe to say aid commitments at time period t are based on data of time period t-1. For that reason all the explanatory variables are lagged one year to better reflect the actual decision-making process. Together with the time-lag between aid commitment and aid disbursement, this measure gives reason to believe a priori that simultaneity will not be an issue. Intuitively, aid to be delivered in the future is not believed to influence past characteristics. An exception is the total amount of ODA allocated by the Netherlands, which is not lagged. This is due to the fact that the aid budget is assumed a given when selection and allocation takes place.
Data
The reliability of statistics from developing countries are often questioned due to, for instance, measurement error. It is probably the case that countries with less developed institutes cannot always provide accurate data. In addition, many observations are simply missing as data was not collected. This is not a significant problem seeing as aid allocation decisions are probably based on information actually available and not on underlying trends not currently recorded McGillivray and White, 1993, p. 58). In line with this, data was
obtained from leading data institutions such as the OECD, World Bank, etc. Since most of the observations display large variability (GDP per capita, ODA commitments etc.) and often skewed distributions (e.g. GDP), the natural logarithm of all variables are taken where possible, dummies excluded. This makes interpretation easier as in a log-log model coefficients can be interpreted as elasticities. In addition, the effect of outliers on results is mitigated this way (Younas, 2008, p. 666). All monetary variables are specified in current US dollars.
C. Method
In the literature three main types of modelling approaches have been employed to model the aid allocation problem.
The Tobit Type I model estimates the choice of recipient and the amount of aid allocated simultaneously. An disadvantage of this approach is that the explanatory variables are expected to have exactly the same effect on selection as on the amount of aid received. It is not inconceivable that certain factors might affect both stages with different magnitude or even different signs. The aid equation of a Tobit Type I model is the maximum of zero and a linear model of explanatory variables:
Yijt = max(aXijt + uijt, 0)
Y is aid received, a is a vector of coefficients on X, the vector of explanatory variables, and u is the error term. I, j and t stand for donor, recipient, and year respectively. Greene (2004) shows that introducing fixed effects into tobit models affects the variance estimators making tests of significance of coefficients problematic at best. A possible solution is incorporating random-effects but estimator consistency then requires the effects to be uncorrelated with the explanatory variables. This is an assumption that usually does not hold and correcting for it is a tedious and complicated process (Berthelemy, 2006a, p.182).
The two-stage model separates the selection and allocation of aid. A usual form is a Probit selection equation and a linear allocation equation.
P(Yijt > 0) = F(bZijt + vijt) Yijt = aXij + uijt
Here b is a vector of coefficients on Z, a vector of explanatory variables in the selection stage. V is an error term, independent from u, and F(.) is the standard normal cumulative distribution function. This procedure relies on selection and allocation being independent. Thereby, it introduces potential for selection bias as factors not included might affect both selection and allocation. In addition, the two-stage model suffers from problems similar to the Tobit model in dealing with fixed effects. Fixed effects are problematic in the selection stage and assuming that they only play a role in the allocation equation seems somewhat arbitrary (Berthelemy, 2006a, p.181).
The Heckman procedure is usually similar to the two-part model with the important difference that uijt and vijt are not assumed independent. In the allocation equation the inverse Mill’s ratio obtained from the selection equation is included as a regressor thus correcting for possible selection bias arising due to the interconnected determination of selection and allocation (Greene, 2012, p. 916).
P(Yijt > 0) = F(bZijt + vijt) Yijt = aXijt + cIMRijt + uijt
Where IMR is the inverse Mill’s ratio, the ratio of the probability density function to the cumulative density function of the normal distribution, evaluated at the predicted outcomes. Technically, this is divided by the standard error of the Probit estimation, however, this is usually assumed to be one (Heckman, 1979). Again, only in the allocation equation can fixed effects estimation be employed.
Since early 2000s, with some exceptions (McGillivray and Oczkowski, 1991), all three alternatives have been employed without a single method coming to dominate the
literature. However, in the literature two-part models, with or without a Heckman correction, are predominantly employed as the need for fixed effects is recognized by most. Often, as here, the focus lies on the allocation equation and including fixed effects only there is seen as satisfactory (Berthelemy, 2006a, p.182).
Due to the impossibility of variables affecting selection and allocation with different signs and magnitudes the Tobit type I model will not be considered. Although some recent papers include pooled OLS estimates (Hoeffler and Outram, 2011) as a comparison, this paper will not do so seeing as the use of OLS for analysis of aid allocation has been theoretically refuted (McGillavray, 2003). This leaves the two-stage model and the Heckman procedure. Both models will be estimated. Investigation of the correlation between the error terms of the selection and allocation equation will provide a rationale for which results to put more weight on. A Hausman test will attempt to provide a decision between fixed effects and random effects estimation in the allocation equation. Diagnostic tests will point out whether the data exhibits heteroskedasticity and whether the inclusion of year fixed effects is warranted. The complete models appear as follows:
Two-Stage Model
P(Yit > 0) = F(b0 + b1*GDP Per Capitait + b2*Other OECD Aid Receivedit + b3*Growth Rate Per
Capitait +b4*Political Terror Scale Dummyit + b5*Freedom House Aggregate Index Dummyit +
b6*Total Trade Recipient-Donor As A Percentage Of Dutch Total Tradeit + b7*Colonies
Dummyit + b8*War Dummyit + b9*Population Recipient Countryit + b10*Total ODA
Commitments By Netherlandsit + vit)
Y*it = a0 + a1*GDP Per Capitait + a2*Other OECD Aid Receivedit + a3*Growth Rate Per Capitait
+a4*Political Terror Scale Dummyit + a5*Freedom House Aggregate Index Dummyit + a6*Total
Trade Recipient-Donor As A Percentage Of Dutch Total Tradeit + a7*Colonies Dummyit +
a8*War Dummyit + a9*Population Recipient Countryit + a10*Total ODA Commitments By
Here vit and uit are assumed independent and Y*it is a subset of Yit for which Yit>0.
Heckman Procedure
P(Yit > 0) = F(b0 + b1*GDP Per Capitait + b2*Other OECD Aid Receivedit + b3*Growth Rate Per
Capitait +b4*Political Terror Scale Dummyit + b5*Freedom House Aggregate Index Dummyit +
b6*Total Trade Recipient-Donor As A Percentage Of Dutch Total Tradeit + b7*Colonies
Dummyit + b8*War Dummyit + b9*Population Recipient Countryit + b10*Total ODA
Commitments By Netherlandsit + vit)
Y*it = a0 + a1*GDP Per Capitait + a2*Other OECD Aid Receivedit + a3*Growth Rate Per Capitait
+a4*Political Terror Scale Dummyit + a5*Freedom House Aggregate Index Dummyit + a6*Total
Trade Recipient-Donor As A Percentage Of Dutch Total Tradeit + a7*Colonies Dummyit +
a8*War Dummyit + a9*Population Recipient Countryit + a10*Total ODA Commitments By
Netherlandsit + a11*Inverse Mills Ratio From Selection Equationit + uit
Here vit and uit are no longer assumed independent and Y*it is again a subset of Yit for which
Yit>0. The inverse mills ratio is given by E[Yit│Yit>0] = μ + σ[ϕ(α – μ)/σ]/[1 – φ(α – μ)/σ],
where μ is the mean of normally distributed variable Yit, σ2 is its variance, α is a constant, ϕ
is the standard normal density function and φ is the standard normal cumulative
distribution function. E[Yit│Y>0] denotes the expectation, or mean, of variable Yit conditional
IV. Analysis and Results
A. Descriptives
Some brief descriptives will be offered to get a rough feeling for the data under consideration. As mentioned the data spans the period 1981-2010, where available.
Variable label explanation and additional descriptive statistics are located in appendix C. A
notable characteristic of Dutch official development aid is that it is spread over many countries. As can be seen from figure 4.1, the ten recipient nations receiving the most ODA from the Netherlands, on average, over the last thirty years together comprise less than half of the total ODA allocated on average. In addition, the Netherlands has allocated aid to 158 developing countries in the last 30 years. This might indicate allocation based more on merit and need as self-interest of the Netherlands is more likely to be represented by a restricted number of nations with which it has special ties. However, one must keep in mind that the effect of taking the average is that large allocations which do not persist, likely an indication of donor characteristics as recipient need and merit are unlikely to change dramatically in short time periods, are averaged out.
Figure 4.1 57.19301081 2.680081651 2.895819145 3.163987133 3.289003448 3.523298051 3.987358968 4.646358323 5.007556029 6.117209011 7.496317435
Percentages of Netherlands' ODA Commitments By
Country
Rest of Developing Nations Burkina Faso Afghanistan
Ghana Suriname Mozambique
Sudan Bangladesh Tanzania
A closer look at the data (figure 4.2), ignoring dummy variables or categorical variables, shows that variables exhibit high variability in range between one another. In addition variances are high, the data are mostly skewed, and mean and median often differ substantially. This indicates that the data is probably not distributed normally. The non-normality of the data is the main motivation for taking the natural logarithm of these variables.
Summary Statistics
Figure 4.2
When looking at the correlation between variables it is apparent that the need variables and population have the strongest correlation with the amount of ODA allocated. The donor characteristics variables have a weak correlation with the dependent variable and the merit variables a very weak correlation. As far as possible multicollinearity is concerned only population potentially poses a problem. The correlation between population and total trade between recipient and donor as a percentage of donor total trade is .6932. The correlation between population and other ODA commitments received by the recipient is .7354. These correlations are strong and intuitively unsurprising as more populous developing nations are expected to receive more aid and trade more in absolute terms than less populous ones. As we will later see, this poses no threat to estimation results. If the dependent variable is specified in per capita terms, and population no longer included as a regressor, the estimates are identical due to the use of logarithms.
Observations 5548 max 422.53 93605.8 20322.02 91.6729 81.7567 1.30e+09 10265.78 min 0 64.8102 0 -50.2358 0 7525 731 range 422.53 93540.99 20322.02 141.9087 81.7567 1.30e+09 9534.78 count 5369 4414 5548 4326 4177 5091 5548 skewness 6.535289 4.775657 12.53399 .7841 13.71965 8.506747 2.102891 variance 647.047 6.02e+07 414995.8 43.15267 12.33058 1.46e+16 3870141 sd 25.43712 7759.545 644.2017 6.569069 3.511492 1.21e+08 1967.267 p50 .09 1343.99 54.385 2.005755 .139978 4600000 2108.62 mean 8.390784 4095.27 232.7313 1.659513 1.041321 2.78e+07 2695.016 ODA GDP OtherODA Growth Trade Population TotalODA .
Correlations
lnODA lnGDP lnOtherODA lnGrowth PTS FH lnTrade Colony Conflict lnPopulation lnTotalODA
lnODA 1 lnGDP -0.382*** 1 lnOtherODA 0.498*** -0.433*** 1 lnGrowth -0.0329 0.0251 0.00969 1 PTS 0.0196 0.0981*** -0.243*** -0.0255 1 FH -0.0110 -0.0849*** -0.140*** 0.0430* 0.117*** 1 lnTrade 0.160*** 0.285*** 0.358*** 0.0395* -0.129*** 0.0765*** 1 Colony 0.172*** -0.00500 -0.0666*** 0.0286 0.0469*** 0.0150 0.122*** 1 Conflict 0.121*** 0.0510*** 0.155*** 0.0992*** -0.0352** -0.0276* 0.159*** 0.301*** 1 lnPopulation 0.441*** -0.341*** 0.689*** 0.0275 -0.232*** 0.0292* 0.640*** 0.0506*** 0.135*** 1 lnTotalODA 0.157*** 0.171*** 0.208*** 0.0362* -0.0702*** -0.168*** -0.0372* -0.00004 0.0469*** 0.0636*** 1
All variables except ODA and TotalODA lagged by one year
* p < 0.05, ** p < 0.01, *** p < 0.001
__________________________________________________________________________________________ Figure 4.3
A look at the scatter plots for correlations between ODA commitments by the Netherlands and the dependent variables shows the presence of a number of outliers (Appendix C). Given that the natural logarithm of the variables is used and the relatively large dataset, 179 countries over 30 years, these outliers are not expected to be influential.
B. Regressions Results and Interpretation
The allocation equation is the focal point of this paper. Before analysing it, we will briefly investigate the selection equation (Figure 4.4, column 1). The surprising finding is that none of the merit variables are significant even at the 10% level and that the only variable signifying donor interest which is significant is the trade variable. According to this estimation the selection of aid recipients is based on recipient need and trade with recipient. As noted, the amount of other ODA received by the recipient could be taken as a proxy for merit instead of need, but seeing as the other three merit variables are not significant this is unlikely. In addition, it is noteworthy that the dummy indicating colonial ties is not significant seeing as this factor is usually found to be significant in the literature (see for instance Alesina and Dolla, 2000), and seeing as Surinam and Indonesia, former Dutch colonies, are among the top ten Dutch ODA recipients. These results should not be overemphasized as we shall see that fixed effects are important in Dutch aid allocation and including them in a Probit regression with consistent estimates is not possible.
Regressions Output Selection Equation and Comparison FE and RE ___________________________________________________________________________ (1) (2) (3) VARIABLES Selection FE RE lnGDP(t-1) -0.561*** -0.874*** -0.872*** (0.0789) (0.108) (0.0883) lnOtherODA(t-1) 0.275*** 0.373*** 0.402*** (0.0457) (0.0538) (0.0498) lnGrowth(t-1) 0.0245 0.0127 0.0115 (0.0373) (0.0335) (0.0333) PTS(t-1) -0.0142 0.128 0.137* (0.0933) (0.0802) (0.0789) FH(t-1) -0.164 -0.0633 -0.0999 (0.124) (0.111) (0.104) lnTrade(t-1) 0.208*** 0.153*** 0.158*** (0.0489) (0.0582) (0.0528) Colony(t-1) 0.586 3.174*** (0.723) (1.054) Conflict(t-1) -0.290 0.297 0.208 (0.479) (0.781) (0.560) lnPopulation(t-1) -0.00548 -0.426 0.107 (0.0673) (0.313) (0.0845) lnTotalODA(t-1) -0.739*** 0.208* 0.0875 (0.0816) (0.107) (0.0794) Constant 10.22*** 10.43** 2.417* (1.315) (4.459) (1.452) Observations 2,451 1,858 1,858 Number of Id 142 135 135 R-squared 0.066 Country FE Yes
Standard Errors in Parentheses *** p<0.01, ** p<0.05, * p<0.1
___________________________________________________________________________
Figure 4.4
For the allocation equation, columns 2 and 3 of figure 4.4. show the estimates using fixed effects and random effects respectively. Fixed effects allow the inclusion of omitted, potentially unobserved, variables that differ among countries but are invariant over time. In practice, this amounts to the same as including a dummy for each country and running a
regular linear estimation. A downside for our case is that a degree of freedom is removed for each dummy included, in this case each country. Random effects include the omitted factors of fixed effects, but also allows for factors which are invariant between countries, but change over time. Fixed effects regression with panel data will always yield consistent estimates, but random effects estimation generally produces more efficient estimators. The established way to choose between the two is a Hausman test, which checks the consistency of the random effects estimators. The most demanding assumption of the Hausman test is that one of the two estimators must be efficient (Woolridge, 2013, p.478) Given that we assume the model to be well specified, and that the sample is relatively large, the Hausman test is expected to provide a reasonably good indication of which estimates to use. Although the estimation results are similar in many points, the Hausman test rejects the random effects model even at the .01% significance level (Appendix C, Figure C.7). This leads us to further investigate only the fixed effects estimates.
The choice between the two-stage model and the Heckman procedure depends on whether the selection equation and the allocation equation are independent or not. The correlation between the residual of the selection equation and the residual of the allocation equation is 42.43% (Appendix C, figure C.8). This seems to warrant the use of the Heckman procedure. Although the Heckman procedure will be posited as the preferred model, the two-stage estimates will be included for completeness’ sake.
To check for heteroskedasticity, the modified Wald test for groups wise heteroskedasticity is computed (Greene, 2012, p.596). Due to the size of the sample under consideration, the power of the test is assumed to be guaranteed. The results point towards the presence of heteroskedasticity (Appendix C, Figure C.9). In addition, to see if time fixed effects are warranted a regression with dummies for each year was performed. The joint test for all the dummies rejects the null hypothesis that all the coefficients are 0 (Appendix C, Figure C.10). Therefore, time fixed effects are included. The individual years’ output is suppressed in figure 5.2. The residuals of the allocation equation are required to be normally distributed to ensure that hypothesis tests yield the right results and p-values are accurate, among others.
A histogram of the residuals shows that the distribution of the residuals matches the normal distribution quite closely (Appendix C, Figure C.11).
Column 1 of Figure 4.5 shows the results of the Heckman procedure. Surprisingly, the only significant coefficients are the coefficients on recipient need variables. Given the log-log specification the coefficients can be interpreted as elasticites. In this case a 10% increase in total ODA commitments received by the recipient nation, Netherlands ODA commitments excluded, results in a 3.4% increase of Dutch ODA commitments towards that recipient, on average. A 10% increase in GDP per capita in a recipient nation results in a decrease of Dutch ODA commitments by 7.15%, on average. The recipient merit variables and donor characteristics are not significant. This is surprising seeing as the Dutch official statement regarding development aid indicates that merit variables are important, and the consensus previously reached in the literature is that self-interest plays an important role for aid allocation decisions of the average country. In addition, trade relationships are stressed by the Netherlands but this is not backed up by the evidence provided here. The fixed effects estimation does imply that specific relations between donor and individual recipients are important, like colonies, but does not exactly quantify this or expose exact determinants. One of the most important requirements of the Heckman procedure is met, the trade variable is significant in the selection equation but not in the allocation equation. Overall, these results indicate that the Netherlands is a very altruistic donor.
The fixed effects estimates in column 2, Figure 4.5, indicate very similar results. Thus although the correlation of the residuals of the selection and allocation equation seems to imply that the two stages are not independent, assuming that they are would only alter the results very marginally.
To check the robustness of the results several alternative specifications have been used. Per capita growth rates were replaced by absolute growth rates and the dummies for the political terror scale and freedom house indices were split at the mean instead of the median. The percentage of trade was replaced by the absolute value of trade between
donor and recipient, and the dummies indicating colonial ties and conflicts were replaced by the number of years of the colonial relationship or conflict duration.
Regression Output Allocation Equation
___________________________________________________________________________ (1) (2) VARIABLES Heckman FE lnGDP(t-1) -0.715*** -0.820*** (0.208) (0.195) lnOtherODA(t-1) 0.340*** 0.392*** (0.103) (0.0944) lnGrowth(t-1) 0.00868 0.0157 (0.0376) (0.0369) PTS(t-1) 0.126 0.131 (0.0985) (0.0987) FH(t-1) -0.0640 -0.0793 (0.155) (0.158) lnTrade(t-1) 0.0818 0.117 (0.110) (0.104) Conflict(t-1) 0.366 0.213 (0.236) (0.218) lnPopulation(t-1) 0.592 0.785 (0.828) (0.790) lnTotalODA -0.0979 -0.267 (0.309) (0.252) Inverse Mills -0.614 (0.549) Constant -4.076 -5.483 (11.51) (11.35) Observations 1,858 1,858 R-squared 0.121 0.119 Number of Id 135 135
Country FE Yes Yes
Year FE Robust SEs Yes Yes Yes Yes Robust Standard Errors in Parentheses
*** p<0.01, ** p<0.05, * p<0.1
___________________________________________________________________________
Due to the presence of many zeroes, the colonial and conflict variables were included in absolute form instead of their natural logarithm. The results (Appendix C, Figure C.12) are largely similar to the results using the original specification. Except for some minor magnitude differences, the only difference is that the political terror scale is weakly significant (at 10%) in the Heckman equation. Alternatively, the original model was estimated twice with half the data, split by country id. This approach is expected to possibly exhibit weaker results due to reduced sample size but might shed some light on the robustness of the results. The results (Appendix C, figure 13) show some differences from the original results, mainly in significance levels. The selection equation of the first half of the data shows weak significance (at 10%) of the trade variable and the freedom house dummy. In the allocation equation the amount of other ODA received is now significant at 5% and, surprisingly, the conflict dummy is significant at the 10% level. This is probably due to the influence of Afghanistan, who has received major amounts of ODA since its invasion in which the Netherlands took part from 2006 onwards. Regression results from the second half of the data are virtually the same in terms of significance in all but one surprising aspect, GDP per capita is no longer significant in the allocation equation.
To address the problem of potential multicollinearity between population and other explanatory variables found in by examining the correlations a new dependent variable was specified. Because the natural logarithm of the dependent variable and population are taken, including population as a regressor and choosing absolute ODA commitments as dependent variable is equivalent to taking ODA commitments per capita as the dependent variable and excluding population as a regressor. However, due to missing values in the data on population the estimation results differ somewhat in magnitude (Appendix C, Figure C.14). In addition, the conflict dummy is now significant at 10% in the allocation stage. These results indicate that multicollinearity between population and trade or other ODA commitments are not a threat to the overall results. Finally, separate estimates were obtained for the periods 1981-1995 and 1996-2010 to see if allocation motives shifted (Appendix C, Figure C.15). For the first 15-year period the most notable differences are that the trade variable is now only significant at the 5% level, and that the freedom house
dummy is also significant at the 5% level. The negative sign of the freedom house coefficient is unsurprising if this variable is seen as an indication of recipient merit. It implies that political and social instability of a nation influences the chances of receiving development aid negatively. The allocation equation returns more surprising differences as the only significant coefficients are population (at 10%) and other ODA commitments received (at 5%). It should be noted that the R-squared for this estimation is low even by panel data standards indicating that this model might not be a good fit to the data. The results for the period 1996-2010 are slightly more in line with expectations although the trade variable is not significant in either equation and the amount of other ODA commitments received is no longer significant in the second stage. Both of the controls are significant at the 5% level. These separate estimations based on time periods might be interpreted as aid allocation decisions shifting from following what other countries have done to looking more closely at recipient need oneself, although the reduced sample size and low R-squared in the first period do not make for solid estimation results.
Overall, these results seem to indicate that the Netherlands is a strongly altruistic aid allocator. In the selection stage there is evidence that countries with strong commercial ties are preferred. In contrast with the literature no “small country effect” is found as the population coefficient is not significant in the allocation equation. Variations in specifications indicate that factors of merit and self-interest might play a modest role but estimation results are not sufficiently clear on this point. The relevance of fixed effects indicates that country specific ties, which might remain unobserved, are important. Estimates of the GDP-per-capita elasticity of ODA commitments lie between -.7 and -.8 whereas estimates of the other ODA commitments elasticity of ODA commitments lie between .3 and .4. In its aid allocation decisions the Netherlands appears to be driven strongly by recipient need and not so much by other factors.
V. Conclusion
This paper has attempted to analyse the aid allocation process of the Netherlands for the period 1981 through 2010. The official statement of the Dutch ministry of foreign affairs indicates that the Netherlands wants to help impoverished nations overcome their problems and eventually turn them into economic partners. This implies aid allocation based on recipient need and recipient merit. However, the consensus achieved in the professional literature is that donor interest plays an important role in aid allocation.
This research finds that the Netherlands is in fact a fairly altruistic aid allocator who focuses mainly on recipient need. Robustness checks identify that certain recipient merit and self-interest variables might be influential in some cases, but the evidence is limited. Given the goals of Dutch official development assistance it seems that the process of aid allocation might be sub-optimal seeing as ignoring recipient merit could potentially result in nations not generating growth from the development assistance (Burnside & Dollar, 2000). Shifting the focus to allocation based on recipient need and recipient merit would help to better achieve the stated goals.
Two main shortcomings of this research are to be noted. Firstly, certain papers have argued that the bureaucratic side of aid allocation is a side needing consideration. These so-called incremental studies focus more on the rigidity of decision making where inertia could be an important aspect of aid allocation. This paper ignores bureaucratic structures, personal and other social characteristics of the actors involved in the process, and potential aid allocation inertia. Secondly, this paper has ignores the potential involvement of the recipient in the aid allocation process. Aid allocation is assumed to be an independent action by the donor based on given information. It is not unthinkable that recipients influence the donor’s decision making process by for instance rent-seeking techniques. While fixed effects estimation could potentially capture some of these factors, a different modelling approach might be required
if such effects are strong and persistent. These two aspects of aid allocation are potentially relevant and should be investigated in detail by future research.
VI. Bibliography
Aid to Developing Countries Rebounds in 2013 to Reach an All-Time High. (2014). Retrieved June 8, 2014: http://www.oecd.org/newsroom/aid-to-developing-countries-rebounds-in-2013-to-reach-an-all-time-high.htm
Aid to Poor Countries Slips Further as Governments Tighten Budgets. (2013).
Retrieved June 8, 2014:
http://www.oecd.org/dac/stats/aidtopoorcountriesslipsfurtherasgovernments tightenbudgets.htm
Alesina, A., & Dollar, D. (2000). Who Gives Foreign Aid to Whom and Why? Journal of
Economic Growth, 5(1), 33-63.
Barbieri, K., & Keshk, O. (2012). Correlates of War Project Trade Data Set Codebook,
Version 3.0. Retrieved May 30, 2014: http://correlatesofwar.org.
Barbieri, K., Keshk, O., & Pollins, B. (2009). “TRADING DATA: Evaluating our Assumptions and Coding Rules.” Conflict Management and Peace Science, 26(5), 471-491.
Berthélemy, J. C. (2006a). Bilateral Donors’ Interest vs. Recipients’ Development Motives in Aid Allocation: Do All Donors Behave the Same? Review of Development
Economics, 10(2), 179-194.
Berthélemy, J. C. (2006b). Aid Allocation: Comparing Donors’ Behaviours. Swedish
Economic Policy Review, 13, 75-109.
Berthélemy, J. C., & Tichit, A. (2004). Bilateral Donors’ Aid Allocation Decisions – A Three-Dimensional Panel Analysis. International Review of Economics and Finance, 13(3), 253-274.
Burnside, C., & Dollar, D. (2000). Aid, Policies, and Growth. The American Economic
Review, 90(4), 847-868.
Clist, P. (2011). 25 Years of Aid Allocation Practice: Whither Selectivity? World
Development, 39(10), 1724-1734.
Correlates of War 2 Project. (2014) Colonial/Dependency Contiguity Data,
Dollar, D., & Levin, V. (2006). The Increasing Selectivity of Foreign Aid, 1984-2003.
World Development, 34(12), 2034-2046.
Dool, Pim van den. (2012). VVD Wil Budget Ontwikkelingshulp Met Zeventig Procent
Verlagen. Retrieved June 8, 2014:
http://www.nrc.nl/nieuws/2012/06/16/vvd-wil-budget-ontwikkelingshulp-met-zeventig-procent-verlagen/
Freedom House (2014). Freedom in the World 1973-2014. Retrieved June 1, 2014: http://www.freedomhouse.org/report-types/freedom-world#.U5Wagvnhs0c
Gang, I. N., & Kahn, H. A. (1990). Some Determinants of Foreign Aid to India, 1960-85. World Development, 18(3), 431-442.
Gibney, M., Cornett, L., Wood, R., & Haschke, P., (2014) Political Terror Scale
1976-2012. Retrieved June 1, 2014: http://www.politicalterrorscale.org/
Greene, W. (2012). Econometric Analysis (7th). Essex: Pearson Education Limited.
Greene, W. (2004). The Behaviour of the Maximum Likelihood Estimator of Limited Dependent Variable Models in the Presence of Fixed Effects. Econometrics Journal, 7, 98-119.
Harrigan, J., & Wang, C. (2011). A New Approach to the Allocation of Aid Among Developing Countries: Is The USA Different from the Rest? World Development, 39(8), 1281-1293.
Heckman, J. C. (1979). Sample Selection Bias as a Specification Error. Econometrica,
47(1), 153-161.
Hoeffler, A., & Outram, V. (2011). Need, Merit, or Self-Interest – What Determines the Allocation of Aid? Review of Development Economics, 15(2), 237-250.
Ketz, L. B. (Eds.). (2002). Encyclopedia of American Foreign Policy: Volume I. New York: Charles Scribner’s Sons.
Levitt, M. S. (1968). The Allocation of Economic Aid in Practice. The Manchester
School of Economics and Social Studies, 36(2), 131-147.
Maizels, A., & Nissanke, M. K. (1984). Motivations for Aid to Developing Countries. World Development, 12(9), 879-900.
McGillivray, M. (2003). Modelling Aid Allocation. WIDER Discussion Paper No.
McGillivray, M., & Oczkowski, E. (1991). Modelling the Allocation of Australian Bilateral Aid: Two-Part Sample Selection Approach. Economic Record, 67, 147-52.
McGillivray, M., & White, H. (1993). Explanatory Studies of Aid Allocation Among Developing Countries: A Critical Survey. Social Studies Working Paper 148.
McKinley, R., & Little, R. (1979). The US Aid Relationship: A Test of the Recipient Need and the Donor Interest Models. Political Studies, 27(2), 236-250.
Ministry of Foreign Affairs Faces Far-Reaching Budget Cuts. (2013). Retrieved June 8,
2014:
http://www.government.nl/news/2013/09/17/ministry-of-foreign-affairs-faces-far-reaching-budget-cuts.html
OECD Development Assistance Committee (2014). Aid Statistics. Retrieved May 30, 2014: http://www.oecd.org/dac/stats/
Rijksoverheid (2014). Dutch Development Aid Statement. Retrieved June 3, 2014: http://www.rijksoverheid.nl/onderwerpen/ontwikkelingssamenwerking
Sarkees, R. R., & Wayman, F. (2010). Resort to War: 1816 - 2007. CQ Press Retrieved June 1, 2014: http://www.correlatesofwar.org/
Stockholm International Peace Research Institute (2014). Arms Transfers Statistics. Retrieved June 1, 2014: http://armstrade.sipri.org/armstrade/page/trade_register.php
Trumbull, W. N., & Wall, H. J. (1994). Estimating Aid-Allocation Criteria with Panel Data. The Economic Journal, 104(425), 876-882.
Younas, J. (2008). Motivations for Bilateral Aid Allocation: Altruism or Trade Benefits.
European Journal of Political Economy, 24(3), 661-674.
Woolridge, J. M. (2013). Introductory Econometrics: A Modern Approach (5th).
Mason, OH: South-Western CENGAGE Learning.
World Bank (2014). GDP and Population Statistics. Retrieved: June 2, 2014: http://data.worldbank.org/
VIII. Appendices
A. Variable Specification
Colony - Dummy indicating whether recipient nation was or is a colony of the
donor nation (1=yes)
Conflict - Dummy indicating whether donor nation was involved in a conflict
on recipient territory that ended no earlier than 1900
FH - Dummy indicating whether an observation lies above (=1) or below
(=0) the median of the average of the civil liberties and political rights indices
FH2 - Dummy indicating whether an observation lies above (=1) or below
(=0) the mean of the average of the civil liberties and political rights indices
GDP - GDP per capita in current US dollars
Growth - GDP per capita growth rate in annual percentages
GrowthRate - GDP growth rate in annual percentages
ODA - Total value of Dutch ODA commitments to recipient in current US
dollars (millions)
OtherODA - Total value of ODA commitments, excluding the Netherlands, in
current US dollars (millions)
Population - Total Population
PTS - Dummy indicating whether an observation lies above (=1) or below
(=0) the median of the Political Terror Scale
PTS2 - Dummy indicating whether an observation lies above (=1) or below
(=0) the mean of the Political Terror Scale
TotalColony - Total amount of years recipient nation was a colony of donor nation,
recipient territory since 1900
TotalODA - Total value of Dutch ODA commitments to all recipients in current
US dollars (millions)
TotalTrade - Total value of trade between recipient and donor, in current US
dollars
Trade - Total trade between donor and recipient as a percentage of donor
GDP
The inclusion of “ln” before a variable name specifies that the natural logarithm was taken. The inclusion of “(t-1)” after a variable name specifies the value of the variable is lagged on period, in this case one year.
B. Countries and Codes
A list of all recipient nations included, and their country codes as defined by the World Bank.
Country Name World Bank ID
Albania ALB
Afghanistan AFG
Algeria DZA
Angola AGO
Anguila AIA
Antigua and Barbuda ATG
Argentina ARG Armenia ARM Aruba ABW Azerbaijan AZE Bahamas BHS Bahrain BHR Bangladesh BGD
Belarus BLR Belize BLZ Benin BEN Bermud a BMU Bhutan BTN Bolivia BOL Bosnia-Herzegovina BIH Botswana BWA Brazil BRA Brunei BRN
Burkina Faso BFA
Burundi BDI
Cambodia KHM
Cameroon CMR
Cape Verde CPV
Cayman Islands CYM
Central African Rep. CAF
Chad TCD Chile CHL China CHN Chinese Taipei CHT Columbia COL Comoros COM
Congo Dem. Rep. ZAR
Congo Rep. CQG
Cook Islands COK
Costa Rica CRI
Cote d’Ivoire CIV
Croatia HRV
Cuba CUB
Cyprus CYP
Djibouti DJI
Ecuador ECU Egypt EGY El Salvador SLV Equatorial Guinea GNQ Eritrea ERI Ethiopia ETH Falkland Islands FLK Fiji FJI
French Polynesia PYF
Gabon GAB Gambia GMB Georgia GEO Ghana GHA Gibraltar GIB Grenada GRD Guatamala GTM Guinea GIN Guinea-Bissau GNB Guyana GUY Haiti HTI Honduras HND
Hong Kong, China HKG
India IND Indonesia IDN Iran IRN Iraq IRQ Israel ISR Jamaica JAM Jordan JOR Kazahkstan KAZ Kenya KEN Kiribati KIR Korea KOR
Kosoco KSV Kuwait KWT Kyrgyz Republic KGZ Laos LAO Lebanon LBN Lesotho LSO Liberia LBR Libya LBY Macao MAC Madagascar MDG Malawi MWI Malaysia MYS Maldives MDV Mali MLI Malta MLT Marshall Islands MHL Mauritania MRT Mauritius MUS Mayotte MAY Mexico MEX
Micronesia, Fed. States FSM
Moldova MDA Mongolia MNG Montenegro MNE Montserrat MSR Morocco MOR Mozambique MOZ Myanmar MMR Namibia NAM Nauru NRU Nepal NPL
Netherlands Antilles ANT
Niger NER Nigeria NGA Niue NIU Northern Marianas MNP Oman OMN Pakistan PAK Palau PLW Panama PAN
Papa New Guinea PNG
Paraguay PRY Peru PER Philiphines PHL Quatar QAT Rwanada RWA Samoa WSM
Sao Tome & Principe STP
Saudi Arabia SAU
Senegal SEN
Serbia SRB
Seychelles SYC
Sierra Leone SLE
Singapore SGP
Slovenia SVN
Solomon Islands SLB
Somallia SOM
South Africa ZAF
South Sudan SSD
Sri Lanka LKA
St. Helena SHN
St. Kitts-Nevis KNA
St. Lucia LCA
St. Vincent & Grenadines VCT
Swaziland SWZ Syria SYR Tajikstan TJK Tanzania TZA Thailand THA Timor-Leste TMP Togo TGD Tokelau TKL Tonga TON
Trinidad and Tobago TTD
Tunisia TUN
Turkey TUR
Turkmenistan TKM
Turks and Caicos Islands TCA
Tuvalu TUV
Uganda UGA
Ukraine UKR
United Arab Emirates ARE
Uruguay URY
Uzbekistan UZB
Vanautu VUT
Venezuela VEN
Vietnam VNM
Virgin Islands VIR Wallis & Futuna WLF West Bank & Gaza Strip WBG
Yemen YEM
Zambia ZMB
C. Descriptives, Tests and Regression Output
C.1 Scatter Plots ______________________________________________________ ______________________________________________________ Figure C.1 ______________________________________________________ _______________________________________________________ Figure C.2_________________________________________________________ Figure C.3
_________________________________________________________
_________________________________________________________
_________________________________________________________
Figure C.5
_________________________________________________________
_________________________________________________________
Hausman Test For Fixed or Random Effects Estimation
Coefficients
Fixed Effects Random Effects Difference Standard Error _____________________________________________________________________________________ lnGDP(t-1) -0.87425 -0.87242 -0.00183 0.061634 lnOtherODA(t-1) 0.373153 0.402084 -0.02893 0.020554 lnGrowth(t-1) 0.012704 0.011529 0.001175 0.003864 PTS(t-1) 0.127696 0.137222 -0.00953 0.014015 FH(t-1) -0.06329 -0.09989 0.036603 0.038428 lnTrade(t-1) 0.153458 0.157592 -0.00413 0.024452 Conflict(t-1) 0.297362 0.207768 0.089594 0.5444 lnPopulation(t-1) -0.42573 0.106995 -0.53272 0.301582 lnTotalODA 0.208231 0.087499 0.120732 0.071392 Constant 10.43223 2.41691 8.01532 4.215976 _____________________________________________________________________________________ Fixed Effects Estimates Consistent under H0 and H1
Random Effects Estimates Efficient under H0, Inconsistent under H1 Test: H0: Difference in Coefficients is Not Systematic
Chi Squared = 90.45 Probability = 0.0000 Figure C.7
Correlation of Residuals
Selection Residual Allocation Residual
___________________________________________________________
Selection Residual 1.0000
Allocation Residual .4243 1.0000
___________________________________________________________
Modified Wald Test for Groups Wise Heteroskedasticity
_____________________________________________________
Chi-Squared = 6035.31
Probability = 0.0000
_____________________________________________________ H0: Sigma(i)2 = sigma2 for all i
_____________________________________________________
Figure C.9
Significance Test for Individual Year Effects
______________________________________________________________ [1] 1981.year = 0 [2] 1982.year = 0 [3] 1983.year = 0 [4] 1984.year = 0 [5] 1985.year = 0 [6] 1986.year = 0 [7] 1987.year = 0 [8] 1988.year = 0 [9] 1989.year = 0 [10] 1990.year = 0 [11] 1991.year = 0 [12] 1992.year = 0 [13] 1993.year = 0 [14] 1994.year = 0 [15] 1995.year = 0 [16] 1996.year = 0 [17] 1997.year = 0 [18] 1998.year = 0 [19] 1999.year = 0 [20] 2000.year = 0 [21] 2001.year = 0 [22] 2002.year = 0 [23] 2003.year = 0 [24] 2004.year = 0 [25] 2005.year = 0 [26] 2007.year = 0 [27] 2008.year = 0 [28] 2009.year = 0 [29] 2010.year = 0 ________________________________________________________________
___________________________________________________________________ 2006 Omitted Due to Collinearity
H0: All Coefficients On Individual Years Are Zero
F(29, 134) = 3.68
Probability = 0.0000
___________________________________________________________________
Figure C.10
Residual Plot for Allocation Equation
______________________________________________________________________
_______________________________________________________________________
C.3 Regression Output
Alternative Specification
________________________________________________________________________
(1) (2)
VARIABLES Selection Heckman
lnGDP(t-1) -0.472*** -0.721*** (0.0780) (0.212) lnOtherODA(t-1) 0.274*** 0.348*** (0.0427) (0.102) lnGrowthRate(t-1) -0.00336 0.0845 (0.0407) (0.0789) PTS2(t-1) 0.0327 0.161* (0.0832) (0.0943) FH2(t-1) 0.0406 -0.00597 (0.116) (0.143) lnTotalTrade(t-1) 0.124*** 0.0905 (0.0477) (0.111) TotalConflict(t-1) -0.174 -0.130 (0.142) (0.0938) lnPopulation(t-1) 0.0568 0.615 (0.0659) (0.834) lnTotalODA -0.817*** -0.162 (0.0716) (0.329) Inverse Mills -0.525 (0.542) TotalColony(t-1) 0.00793 (0.00574) Constant 8.248*** -4.688 (1.258) (11.60) Observations 2,895 1,855 R-squared 0.123 Number of Id Country FE Year FE Robust Ses 143 135 Yes Yes Yes Standard Errors in Parentheses *** p<0.01, ** p<0.05, * p<0.1
___________________________________________________________________________
