Reshaping geographical allocation of aid: the role of immigration in Spanish Official Development Assistance

The intensification of the international migration flows sent by developing countries and received by developed countries is ‘reshaping’ the aid map of the 21st century. However, to date the influence of immigration on the geographical distribution of aid has been little studied. This paper proposes a general framework for analysing the extent to which immigration flows affect the allocation of Official Development Assistance (ODA). We apply this model to the case of Spain (a country that, in a short time, became one of the top-ten bilateral donors and one of the main recipients of immigration in Europe) during the period from 1998 to 2009 (prior to the current economic crisis, which has slowed down the immigration flows temporarily and reduced the ODA budget drastically). The estimations reveal that immigration is relevant both to the selection of aid partners, and to the allocation of aid quotas, thus ‘reshaping’ the geographical strategy of Spain’s public aid.


Introduction
The allocation of international public aid should be coherent with the officially proclaimed international development agenda. Although few policymakers would deny this assertion, the truth is that the debate on the 'geopolitics of international aid' has been in force since the beginning of the cooperation system. Just a decade after the launch of the first aid programmes, the leading economists such as the Nobel laureate Gunnar Myrdal warned that the only way to allocate resources in a purely philanthropic way was to give up the 'bilateralism' of the aid system and to delegate the management of the resources to a single multilateral agency (Myrdal 1956: 124). Indeed, if the international donor community shared the same 'altruistic' motivations, it would probably be enough to manage aid through a single multilateral agency.
Nevertheless, the prolific literature on the geographical distribution of Official Development Assistance (ODA) suggests that donors do not allocate aid for purely altruistic reasons, and neither are they particularly consistent with their international development commitments. In reality, aid is distributed in a fairly 'eclectic' way, so that developing countries with greater political, historical and cultural affinities with donors, as well as countries with greater economic and geostrategic importance, receive more aid than other countries with similaror greaterlevels of need. As a result, this lack of coherence between the overall objective of aid ('promoting international development') and the criteria that guide the distribution of resources (a mix of altruistic and partisan motivations) cuts down the efficiency of aid and represents a 'constituent defect' that is difficult to eradicate.
Despite the strength of this debate, it is surprising that researches have paid limited attention to one of the most dynamic variables of the current globalisation process: the migratory flows originating in developing countries and received by (developed) donor countries, which have been of great concern to the latter during the so-called 'age of migration' (Castles and Miller 2003). In this sense, we wonder whether or not the reception of immigrants from countries that are not linked to the donor historically is changing the definition of the traditional foreign policy priorities, thus affecting ODA policies, which can be used to support the donors' immigration interests.
The case of Spain is particularly interesting, as immigration has been a phenomenon of great demographic and economic importance there since the 1990s. 1 In a short time, Spain became one of the top-ten OECD donor countries and one of the main recipients of immigration in Europe (in 2010, the foreign population accounted for 12.3 per cent of the total Spanish population and a half of these immigrants came from developing countries). 2 In response to this reception of immigrants, Spain added some of these 'emigrating countries' to her traditional list of aid partners (mostly, countries with historical and cultural links with Spain such as Equatorial Guinea, the Philippines, Morocco and Latin American countries). Consequently, the Spanish aid map expanded as the number of 'prioritised' countries increased.
The aim of this paper is to analyse the extent to which the intensification of migration flows from developing countries to OECD donor countries can change the geographical allocation of foreign aida change that may be affecting especially those countries with large immigration populations, such as France, the United Kingdom, Germany, the USA and Spain. For this purpose, we propose a general framework of analysis, and we apply it to the Spanish ODA allocation case, in the period from 1998 to 2009 (prior to the current economic crisis, which has slowed down the immigration flows temporarily and reduced the ODA budget severely). The paper is structured as follows: in the first section, we briefly review the aid allocation literature and the (neglected) link with international migrations. The second section develops the analytical model of aid allocation and migrations. In the third section, we apply this model to the specific case of Spain. Finally, we summarise the main conclusions of the analysis and propose economic policy criteria to integrate the relation between aid and immigration in a more efficient geographic strategy of aid allocation.

The aid allocation literature and the (neglected) link with international migrations
The analysis of geographical allocation of aid started in the late 1960s with the aim of understanding and improving the distribution of the scarce international resources devoted to the progress of developing countries. 3 The early studies tried to establish the methodological bases for the analysis of the variables that determine the donors' distributive patterns, thus developing the theoretical framework of the initially denominated recipients' needs vs donor's interests approach. Nevertheless, this dichotomy between a 'developmental' model of aid allocation and an 'instrumental' one did not fit in with the complex reality of the donors' decisions. This is the reason why most recent models have assumed that donor Governments distribute their resources according to different variablesgiven a predetermined aid budget. Three explanatory factors have been identified: (i) the relative needs of recipient countries; (ii) the economic and geostrategic interests of donors; and (iii) the recipient countries' capacity to use aid efficiently.
These 'integrated' models of aid allocation adopt the following general expression: where A t is the donor's aid budget in period t; N j is a vector of variables that explains the level of necessity of the recipient country j; I j is a vector of variables that explains the donor's interests in the recipient country j; and G j is a vector of determining variables of aid effectiveness. Despite the abundant empirical literature generated in the last five decades, few authors have developed theoretical models of geographical allocation of aid. The seminal contribution of Dudley and Montmarquette (1976) adopts a microeconomic approach, considering the donor's aid as a good that is consumed indirectly by its citizens. The basic assumption is that 'people usually give because they expect to get something in return' (Dudley and Montmarquette 1976: 133). Thus, donors expect that aid partner countries will support their foreign policies, besides the altruistic 'satisfaction' they derive from helping developing countries. This approach has given rise to other theoretical models, such as the contributions of Trumball and Wall (1994), Wall (1995), Tarp et al. (1999), Feeny andMcGillivray (2004), andTezanos (2008b).
The early empirical studies used cross-section regression analysis (for example, Levitt 1968;Mckinlay and Little 1977;Maizels and Nissanke 1984), and later contributions adopted increasingly sophisticated econometric models with panel data sets and limited dependent variables (such as Tarp et al. 1999;Alesina and Weder 2002;Neumayer 2003;Berthélemy and Tichit 2004;Berthélemy 2006;Isopi and Mavrotas 2006;Tezanos 2008b;and Szent-Iványi 2012). Overall, the empirical studies have contributed to a better understanding of the patterns of aid giving of different bilateral and multilateral donors, who combine altruistic, instrumental and technical (efficiency) considerations. Although there are important differences among donors, the recent empirical studies offer seven consensual results that depict the pattern of aid allocation in the 21st century (Tezanos 2008a: 26-27): (i) Foreign policy interestsboth economic and geostrategicstrongly influence the aid allocation decisions, although there has been an increasing 'developmental' concern since the adoption of the Millennium Development Goals (MDG). (ii) Middle-income countries receive a 'disproportionate' amount of resourcesrelative to their 'aid needs'due to the donors' economic and political interests, and the historical and cultural links. This trend has become less evident in recent years due to the emphasis of the MDG agenda on the poorest countries. (iii) Aid benefits especially small countriesthe so-called 'small countries bias'. (iv) Bilateral donors give greater aid than multilateral ones to small and relatively more advanced developing countries, due to the donors' foreign policy interestsalthough northern European countries are an exception to this general rule. (v) Some donors believe that the impact of aid depends upon the quality of governance and institutions of the recipient countries. In this way, donors try to link aid allocation and effectiveness, although in practice there are important differences among donors' approaches. This increasing emphasis on governance acts to the detriment of the allocations to the so-called 'difficult partnership countries', where aid is potentially less effective. (vi) Donors allocate aid in a 'gregarious' way, which is a consequence of the lack of geographical coordination that generates the existence of aid under-and overallocated countriesthe so-called 'aid darlings' and 'aid orphans'. (vii) At the same time that the MDG agenda invigorates the 'slogan' of poverty reduction, there is a resurgence of other considerations related to 'international security', which explains the high aid allocations to countries such as Afghanistan and Iraq.
Overall, despite this relatively long and rich debate, the literature has paid limited attention to the increasing importance of immigration as a determining factor of geographical allocation of aid. However, in an early study on the historical trends of foreign aid, Hjertholm and White (2000) claim that, since the 1990s, the criteria for bilateral allocation decision have been shifting towards new objectives, such as mitigating the root causes of the heightened migration flows from developing countries to OECD countries. In this line, Xenogiani (2006) in her review for the OECD on the relation between migrations and foreign aidargues that aid is a widely used policy to fight the causes of migration in developing countries, although 'overall the connections between aid, poverty reduction and migration reduction have not been proved empirically' (Xenogiani 2006: 29).
In line with the general literature on aid allocation, researchers have identified three underlyingand not necessarily conflictingpolicy motives for an increased aid support to emigrating countries: first, the 'developmental motive' for supporting the emigrating countries (understanding immigration flows as an additional indicator of the degree of aid need); second, the 'instrumental motive' for supporting the donor's migration management interests by preventing further emigration flows; and third, the 'political motive' for paying attention to the increasing political influence of immigrants' lobbies in the donor country and alleviating the internal tensions that arise from the entry of uncontrolled immigration.
Despite the general neglect of the international migration flows, three notable exceptions have considered specifically the role of immigration in the aid allocation decisions: Lahiri and Raimondos-Møller (2000) have developed a political-economic model of aid allocation based solely on the immigrants' political influence on the donor country. The model is based on the 'political contributions approach' and it assumes that each national-immigrant group in the donor country lobbies the Government to allocate more aid to its country of origin, and the Government accepts political contributions from lobby groups. 4 As the amount of these contributions is positively related to the number of people of each national-immigrant group, the model predicts that the greater the number of immigrants from a specific developing country, the greater the aid that the Government will allocate to this country. Czaika (2005) uses a political-economic framework to model the influence of immigrants (living permanently in the donor country) and asylum seekers (immigrants without permanent residence permits) on the aid allocation pattern of Germany. Using an integrated model of aid allocation (such as the one expressed in equation [1]) for the period from 1991 to 1999, he estimates a positive and significant impact of immigration on the German ODA allocation (with an elasticity of about 0.11). This paper concludes that immigrants 'play a decisive role in forming public opinion, and through this, in shaping the political aid allocation decision-making process of the donor country' (Czaika 2005: 301).
Moreover, Czaika and Mayer (2011) study the influence of 'refugee migration movements' on the allocation of long-term aid and short-term (emergency) aid by 18 OECD donor countries in the period from 1992 to 2003. They distinguish between three types of forced migrants: internally displaced persons who stay in their countries of origin; cross-border refugees that flee to neighbouring countries; and asylum seekers in OECD countries. Their estimates suggest that donors gave emergency aid to all types of forced migrants, but they predominantly allocated long-term aid to the sending countries of asylum seekers. Overall, this study suggests that 'bilateral aid allocation policies are primarily focused on countries of origin [which] indicates that the underlying interests of donor states seems to be more focused on migration prevention than on altruistic burden-sharing motives' (Czaika and Mayer 2011: 468).
Finally, it should be mentioned that the opposite relation between aid and migrations (i.e. how the reception of aid may encourage immigration flows to donor countries) have been even less studied. In this line, Berthélemy et al. (2009) try to fill in this gap investigating the impact of aid on immigration. Assuming the existence of the so-called 'hump-shaped pattern' of migration (i.e. the inverse-U relation between per capita income in the sending countries and their propensity to migrate), 5 the estimation of a model of two simultaneous equations identified a bi-directional interaction between aid and migration. On the one hand, bilateral aid influences migration by enhancing the information about labour market conditions in the destination country (the 'attraction effect'). On the other hand, total aid correlates with migration through increasing wages in countries of origin (the 'push effect'). These authors have computed the per capita income threshold above which emigration and income become substitutes (approximately US $7,300 in PPP 2,000 prices) and they argue that, for countries below this threshold, additional aidprovided that it is effective in stimulating growthwill increase emigration flows.
A general model for understanding the influence of immigration on decisions concerning geographical allocation of aid Following the pioneer contribution of Dudley and Montmarquette (1976), we propose a general framework for analysing the influence of immigration flows on the geographical allocation of a donor country by focusing the analysis on the economic behaviour of the decision-makers responsible for aid allocation. Our model conceives geographical allocation of aid as a two-step decision process, taking into account the censored nature of the dependent variable: given a predetermined annual amount of aid, in the first stage, the donor government chooses a group of partner countries that will receive bilateral aid; in the second stage, the donor determines the aid quotas of each partner country.

First decision stage: selection of aid-partner countries
We analyse the selection stage according to the following attraction index: 6 where Λ jt measures the interest of the donor in the developing country j. N, I, G, H and M are, respectively, vectors of explanatory variables regarding the recipient countries' needs, the donor's interests, the determining factors of aid effectiveness, the allocation's path dependence (history) and the immigration flows received by the donor. Moreover, β, δ, φ, θ and Ψ are sets of weights, all of them constrained within the interval [0, 1] in order to reflect the possible existence of decreasing marginal returns. 7 The parameter α j measures the fixed effects associated with each recipient country, which are not determined by the explanatory variables.
Once the donor estimates the attraction indexes for each partner country, the following selection rule is applied: where D jt = 1 indicates that the country j is chosen as a partner and A t is the predetermined amount of bilateral ODA geographically allocable among J possible developing countries. Recipient countries are ranked according to their respective scores on the attraction indexes, which determine their selection probabilities, so that those above the selection threshold A t ϑ are finally chosen as aid partners. The result of choosing the country j as a partner is thus interpreted as the difference in the utility obtained by the donor between giving and not giving aid to this country, being positive in the event of selection and zero or negative otherwise.
The parameter ϑ reflects the donor's aversion/predisposition to disperse its aid budget among J eligible countries: if ϑ>0, there is aversion to dispersion (as ϑ tends to ∞, the dispersion of the resources is penalised and the selection probability decreases); on the other hand, if ϑ<0, there is predisposition to dispersion (as ϑ tends to −∞, both the dispersion of the resources and the probability of being chosen as an aid partner increase). Also, if ϑ = 0, the selection rule does not depend on the volume of aid.
Substituting [2] into [3] and using logarithms yields the following linear probability function: Hence, we expect the following relations in the process of selecting aid partners: The probability of being selected as an aid partner depends on, simultaneously and positively, the developing country's relative level of need, the interest it has for the donor's foreign policy, the factors determining aid effectiveness, the allocation's path dependence and the immigrations flows received by the donor.

Second decision stage: allocation of aid quotas
Once a subset of K-partner countries has been selected (being K ⩽ J), the donor government decides on the specific quotas of each country in the ODA budget: where a jt measures the share of the country j in the donor's aid and A t is the total amount of bilateral ODA geographically allocable among K-recipient economies: We define the donor's objective function for aid allocation as follows: where the variables are interpreted as in previous equations. Again, the parameters α j , β, δ, φ, θ and Ψ are constrained within the interval [0, 1] so as to indicate the possible existence of decreasing marginal returns. The aim of the donor government is to maximise the total utility derived from the aid allocations to K partner countries, subject to the budget constraint (assuming that the whole aid budget is disbursed): This restriction implies that a marginal increase in the aid quota assigned to a specific partner country will decrease the share of at least one other country. The lagrangian that results from maximising [7] subject to [8] is: The first order conditions are: Working out the value of a jt , we get the aid shares received by each developing country: Taking logarithms in equation [11] yields the linear function: The expected relations in the allocation of the aid quotas are: To sum up, the whole process of aid allocation is structured in two consecutive decisionsgiven a predetermined aid budgetwhich clearly consider the potential influence of immigration flows: first, the donor government chooses a group of aid-partner countries based on the 'attraction' that each country exerts on it (thus computing the selection probabilities described in equation [4]). Second, for those countries previously selected, the donor government distributes the aid budget across the group of partner countries (applying the aid quotas defined in equation [12]). This specification allows us to estimate both decisions separately and to consider a different set of explanatory variables in each decision stage. 8

Case study: Spain
Equations [4] and [12] establish a general framework for studying the influence of international migrations on a donor's aid allocation decisions. We now apply this general model to the specific case of Spain. For this purpose, we first review the studies on the geographical allocation of Spanish aid. After that, we identify the 'determining factors' of the Spanish aid allocation and explain the estimation procedure. And finally, we present the main results of the estimation.

Studies on the geographical allocation of Spanish aid
Spain's ODA has been unusually concentrated on middle-income countries due to its special orientation towards Latin America and North Africa (the former Spanish colonies). Despite this unusual geographical specialisation, only three pieces of research have studied this donor's pattern of aid giving (Alonso 1999;Sánchez Alcázar 1999;and Tezanos 2008b). Other studies have compared the allocation patterns of different donors, using 'standard' models of analysis (applying the same general model to different bilateral and multilateral donors), although few of them have looked specifically into the case of Spain (such as Alesina and Weder 2002; Berthélemy and Tichit 2004;and Isopi and Mavrotas 2006). 9 Overall, these studies have identified an 'eclectic' pattern of Spanish aid allocation that combines both recipient needs and foreign policy interests, but prioritises its particular economic and cultural links with its former colonies. The Spanish pattern of aid giving shows some similarities with those of the USA, Japan and France (which are also strongly determined by their preferences for their respective geographical interests) and it differs greatly from those of the Scandinavian countries, the Netherlands and Canada (which are, to a larger extent, oriented towards the developing countries with the greatest needs).

Determining factors for Spanish aid allocation
Spain's aid allocation decisions depends on several variables, which, for reasons of simplicity, we group into five explanatory factors: (i) recipient countries' needs, (ii) Spain's foreign policy interests, (iii) aid effectiveness determinants, (iv) allocation's path dependence, and (v) immigration flows received by Spain.
Regarding the recipient countries' relative needs, the Spanish Aid Law (1998) has established that the main objective of aid is to contribute to the 'eradication of world poverty, in all its manifestations' (Article 1) and, therefore, aid should be allocated to the 'less economically and socially developed countries' (Article 5).
With regard to foreign policy interests, the Spanish Aid Law establishes clearly that the geographical priorities are 'Latin American countries, the Arab countries of North Africa and Middle East, as well as other less developed countries that maintain special historical and cultural links with Spain' (Article 6). These historical links are determined by a series of cultural and institutional affinities, such as language, religion and legislative traditionfactors that Spain has considered instrumental to facilitating more effective cooperation relations.
Furthermore, Spain may be stimulating its trade and investment interests through the allocation of aid to its main economic partners. In fact, this was the original aim of the Development Assistance Fund, which was set up in 1976 in order to foster the exports of Spanish enterprises by means of tied concessional loans to developing countries.
Regarding the determining factors of aid effectiveness, the international community has increasingly reached a consensus on the significance of the recipient countries' political and institutional environment in the development process and the effectiveness of aid. 10 Since 1989, good governance has been one of the main concerns of the Development Assistance Committee (DAC), which points out the existence of a 'vital connection between open, democratic and accountable political systems, individual rights and the effective and equitable operation of economic systems '. 11 In particular, Spanish aid should promote good policies, as it is expressly ratified in the Aid Law.
With regard to the aid's path dependence, the 'inertial effect' exerted by previous allocations has several explanations: • Donors tend to cooperate with a steady group of partner countries so as to minimise aid administration costs (because adding new partners means incurring additional expenses). Moreover, the donor's management may benefit from 'learning economies', based on previous experiences of assisting the same group of countries, which lead to increased administrative efficiency levels. Spain's effort to acquire greater administrative capacity in its 'priority countries' determines, to some extent, these countries' future probabilities of being aid recipientsand, in the end, their permanence on the list of priorities. 12 • The donor is responsible for providing its partners with stable assistance in order to generate long-term sustainable development processesprovided that the terms of cooperation are fulfilled. In this way, Spain is emphasising the use of programme aid (instead of short-term projects) and new aid instruments (such as global funds, budget support and sector-wide approaches), which have longerterm perspectives. • The inertia is led by previous experiences where aid was shown to be particularly effective. • Long-running political commitments between Spain and recipients have an outstanding importance in the allocation process and contribute to an additional factor of stability, as it is especially complicated to retract resources from a longrunning aid partner.
As regards the immigration flows received by Spain and sent by developing countries, the Spanish aid response to immigration flows may be supporting the three different motivations previously explained ('developmental', 'instrumental' and 'political' motives). 13 In this way, new demand factors (greater importance of the aid needs of those countries that are sending migrants to Spain) and supply factors (Spain's own instrumental and political interests, as aid donor and immigrant recipient country) are taking shape, and affecting the pattern of aid giving.
The importance of immigration to the Spanish ODA policy is reflected in the Aid Plan, which includes among its 11 sector priorities a specific sector devoted to migration and development. The Plan's point of departure is that 'the fight against poverty is an end in itself, its cause not immediately related to migration, and a conviction that migration must be the upshot of a free decision, a personal option, and not of need' (MAEC 2009: 23). Following this premise, the Plan states that the general objective of this sector is 'to promote the positive effects between migration and development', and it does not mention the interest of Spain to control the immigration flows. Consequently, the Spanish ODA disbursements should not be linkedat least not officiallyto the immigration policy.
In the case of Sub-Saharan Africa, the link between the Spanish aid policy and the immigration policy has been particularly explicit in recent years. Thus, the two Africa Plans approved to date (for the periods from 2006 to 2008 and from 2009 to 2012) defined the Spanish foreign policy towards this region, including the aid policy. The first Plan identified 27 African countries as particularly relevant to Spain; ten of these countries were also especially important for Spain in terms of the intensity of the immigration flows. 14 The second Africa Plan reduced the number of priority countries to 25, but did not alter the inclusion of those Sub-Saharan countries that were sending migrants to Spain.

Econometric procedure
The estimation of the Spanish aid allocation pattern requires us to tackle the censored nature of ODA allocations (i.e. the exclusion of some developing countries from Spanish assistance implies that aid is partially continuous with positive probability mass at the value of zero). In this sense, aid flows can be represented by the selection of a thresholdwhich is a latent variablewhere the donor starts to disburse positive amounts of aid (see the attraction index defined in equation [2]). If the observations for a jt = 0 are excluded from the sample, the estimates will be biased and inconsistent, as would be the case of an ordinary least square estimation. This is the reason why limited dependent variable regression models do not omit these null observations, allowing the 'latent' decision of excluding those countries that lie under the threshold level to be analysed. 15 Thus, the present analysis uses a two-part model for the estimation.
We estimate the aid selection stage defined in equation [4] by means of the following regression function with a binary dependent variable and a panel data set: 16 [14] where α j , β, δ, φ, θ and Ψ are the parameters to estimate. We estimate the aid share stage defined in equation [12] by means of the following dynamic panel data regression function: Journal of International Relations and Development Volume 19, Number 3, 2016 where the variables are defined as in previous equations and a i, j * represents the potential aid endowments. Furthermore, the model of the second decision stage includes an explanatory variable that is not strictly exogenous (i.e. it is correlated with past or actual realisations of the error term). Consequently, the assumptions of the classic regression model are not satisfied, which leads to biased estimations. This is the case of the aid inertia variable (the lag of the dependent variable), which implies a problem of endogeneity. The way to solve this problem is to apply consistent estimation methods that take into account fixed effects and endogenous independent variables. Instrumental variable models, which replace non-strictly exogenous variables by strictly exogenous instrumental variables, are generally used in these cases; the instruments are correlated with the explanatory variables and turn out to be orthogonal to the error term. Dynamic regression models with panel data are estimated by the generalised method of moments (GMM), proposed by Arellano and Bond (1991) as a particular case of instrumental variable models. The advantage of the GMM is the use of internal instruments that may include lagged values of the non-exogenous regressors, leading to an improvement of the estimation results. We use the system GMM proposed by Arellano and Bover (1995), and Blundell and Bond (1998), instead of the difference GMM initially proposed by Arellano and Bond (1991). The latter transforms the model by doing first differences to remove unobserved fixed-country specific effects, and instruments the non-strictly exogenous explanatory variables by a moment condition's matrix. On the other hand, the system GMM makes up two equations: the original equation in levels and the firstdifference equation. This system, free from correlated fixed effects, allows us to use more instruments and, consequently, improves the efficiency of the estimation. 17 Until now, the system GMM has not been applied to analyse the patterns of geographical allocation of aid.
The model is estimated by the econometric software STATA, with four additional commands that optimise the estimation: 18 (i) White's standard errors, which are robust to arbitrary heteroscedasticity for the same country; 19 (ii) small sample correction for the covariance matrix estimation; (iii) restriction of the matrix of instruments, creating an instrument for each variable and lag distance, rather than an instrument for each period, variable and lag distance; in small sample sizes, this reduces the bias that stems from the fact that the number of instruments approaches (or exceeds) the number of observations; and (iv) two-step estimations, applying Windmeijer's finite samples correction in order to eliminate standard error biases. Finally, in order to check the validity of the instruments matrix in levels, we carry out the Sargan and Hansen hypothesis tests, as well as the Arellano and Bond test for autocorrelation of the idiosyncratic effect (if this kind of autocorrelation exists, the use of lagged values as instruments will be invalidated).

Measurement of the dependent variable: Spanish bilateral ODA
We analyse the bilateral resources classified by the DAC as ODA. As Spain does not report on commitments, the study uses gross aid disbursements. Moreover, emergency aid is subtracted from the ODA disbursements because it is assumed to be disbursed to countries in emergency situations in an 'additive' way (i.e. in addition to the resources that were already allocated to them), its geographical distribution being independent of the previously resolved one.
With regard to debt relief actions, these multilateral programmes impose a specific plan of execution, where individual donors cannot exert direct influence on the geographical orientations. 20 However, debt forgiveness is not strictly exogenous from each bilateral donor's process of allocating aid, as donors could bring forward the amount of resources previously assigned to the countries favoured by debt cancellations. In fact, multilateral debt programmes specify detailed time schedules for the relief flows, which are often negotiated in international forums before bilateral donors decide the allocation of their own resources. Despite this fact, the Spanish allocation has been affected occasionally by major debt actions, 21 partially because debt reliefs are managed by the Ministry of Finance independently of the Ministry of Foreign Affairs and Cooperation. This highly independent management implies a lack of coordination between both ministries, which consequently makes it difficult to anticipate the eventual aid allocation to those countries that benefit from debt relief. Thus, we subtract debt actions from the ODA disbursement as we consider that the resulting amount is a better approximation of the resources finally allocated by aid decision-makers.
Once the Spanish ODA gross disbursements are obtainednetted of emergency aid and debt reliefsthe dependent variable of the aid partners selection stage is computed by means of a dummy variable that classifies developing countries into two possible categories: 'D = 1: selected partner country' and 'D = 0: otherwise'. The classification rule is the existence of a 'significant' disbursement of aid. As McGillivray and Oczkowski (1992) first pointed out, it is convenient to use a 'minimum threshold' of aid receipts to compensate for the limited impact of highly scattered aid allocations that result in a certain number of recipients with particularly low shares. While choosing a specific threshold level can, ultimately, be an arbitrary procedure, it is especially convenient in the case of Spanish ODA, given the high level of geographical dispersion. Therefore, the first-step estimation considers a partner country only if it receives at least a one per cent share of Spain's ODA. Otherwise, the logit regression would be seriously unbalanced, with a greater share of category 1 (i.e. the number of selected countries is greatly superior to the number of non-selected ones), 22 overestimating the probabilities of being selected. 23 We measure the dependent variable of the aid share stage as each partner country's quota in the predetermined annual amount of Spanish aid. This definition in percentage terms has two relevant virtues: it eliminates the bias introduced by the comparison of figures coming from different years, as it is possible to use gross disbursements data in current dollars; and it avoids bias that stems from the fact that the Spanish aid budget steadily increased until 2008.

Measurement of the explanatory variables
The selection of the explanatory variables follows three criteria: first, it includes those variables that best capture the peculiarities of the Spanish ODA policy; second, it is guided by the literature review of previous studies on aid allocation; and third, from a more practical perspective, it tries to both maximise the data available for developing countries (avoiding in this way a sample selection bias due to a nonrandom omission of data), and to avoid informative redundancy (that may cause multicolinearity problems). Table 1 lists the variables included in the analysis and Appendix A shows the descriptive statistics.

Recipient countries' needs
We use the per capita GDP as an indicator of the average economic welfare of the recipient societies. This allows us to test the existence of a 'progressive distributional criterion'. We include recipient countries' population sizes in order to test their sensitivity to the most populated countries. Given the special interest of Spain in its former colonies (which, except for Mexico, the Philippines and Colombia, were countries of fewer than 45 million inhabitants in 2009), we test the existence of a small countries bias (i.e. whether or not the population coefficient is smaller than one).
Other aspects of social well-beingapart from the per capita income levelsare approximated by means of life expectancy at birth. This variable is both less correlated with per capita income than other synthetic indicators (such as the Physical Quality of Life Index and the Human Development Index), therefore reducing the risk of multicolinearity, and is widely available among developing countries, thus reducing the risk of a non-random sample selection bias. Moreover, sharp reductions over time in the life expectancy may reflect an increase in the country's need for aid due to the negative effects of conflicts, natural disasters or diseases.
We use the share of each recipient country in global ODA (i.e. total aid disbursed by multilateral and bilateral donors), excluding Spain, in order to test the level of coordination among donors. In this respect, there are two possible scenarios. On the one hand, a negative relation between the Spanish allocation and the rest of the donors' allocations might reflect a certain level of coordination, avoiding infra and supraallocations in specific countries. If this were the case, it would be reasonable to consider that the Spanish middle-income country orientation is driven by a specialisation pattern in Latin America, 'agreed' upon within an international coordination scheme. On the other hand, a positive relation could reflect the existence of the so-called 'bandwagon effect'. As Dudley and Montmarquette (1976) first explained, this behaviour implies that the donor's perception of the impact of its aid on a specific developing country is positively influenced by the volume of aid that this country attracts from other donors.

Spain's foreign policy interests
The importance of special historical and cultural links is tested by means of a dummy variable, with two categories: Col = 1 if the country was part of the Spanish colonial empire, and Col = 0 otherwise.
We evaluate the trade interests by means of the share of each developing country in Spanish exports, and we measure investment interests by the net stock of Spanish investment in each developing country, accumulated since 1993. It should be pointed out that, given that a small part of the ODA consists of 'tied loans', there is a potential risk of simultaneity between exports and aid disbursements. Nevertheless, exports are lagged one year, which reduces the risk of simultaneity, since tied aid, for the most part, stimulates the current year's exports.

Determining factors of aid effectiveness
In order to approximate the partner countries' commitment to good governance, we use the Worldwide Governance Indicators developed by Kaufmann et al. (2012). In particular, we use the variable of control of corruption, so the coefficient of this variable is expected to be positive, reflecting Spanish support for the least corrupt countries.
Furthermore, the recipient economies' absorptive capacities are included in the model so as to take into account the productive constraints of additional aid allocations and the existence of decreasing marginal returns on aid. We proxy this concept by means of the ratio of the total ODAbilateral and multilateralreceived by the country to its GNI (ODA/GNI ratio), which is widely available among developing countries and offers a reasonable measure of the recipient economy's aid dependency level. 24 Aid's path dependence We include the previous year ODA-quota in both decision stages in order to capture the path dependence of the geographical allocations, assuming that previous disbursements influence positively both the probability of being 're-selected' as an aid partner, and the final amount received.
Immigration flows Immigration flows are measured by the share of foreign population (by nationality) in relation to the total Spanish population (Eurostat 2012).
It should be borne in mind that, despite the fact that Spain has being receiving immigrants from over 130 different developing countries during the period from 1998 to 2009, the immigrant population is concentrated on a limited number of countries. Thus, the top-ten developing countries sending migrants to Spain accounted for 77.5 per cent of the total of immigrants from developing countries (see Appendix B). In particular, two facts are worth noting: on the one hand, Morocco, Ecuador, Colombia and Bolivia are, by far, the most important countries in terms of immigrants living in Spain. And, on the other hand, only three of the top-ten countries during these 11 years do not have historical links with Spain: China, India and Algeria (and, among them, only the Chinese have been consistently a large immigrant community in Spain during the whole period).

Model specification, sample and period of analysis
As equations [4] and [12] specify, we transform the model in a linear function by means of the natural logarithm, thus facilitating the interpretation of the coefficients in terms of elasticities and reducing the heteroscedasticity among observations. 25 In order to access the information available to Spanish aid decision-makers in a realistic way, we specify the explanatory variables with different time lags (Table 1). The lengths of the lags depend on the time delay that takes place in the provision of international statistics: 26 decision-makers faced a two-year information lag in the case of per capita incomes, population levels, life expectancy and absorptive capacities. Immigration flows, aid disbursements, and Spanish exports and investments are lagged one year. This lag structure also reduces the potential simultaneity bias between the aid allocation and the explanatory variables.
The analysis includes the 163 developing countries referred to by the successive DAC lists of ODA recipients. Of these countries, 129 received Spanish ODA for at least one year. Thirty countries were eventually excluded from the analysis due to lack of information. 27 The period of analysis runs from 1998 to 2009, excluding both the first 'gestation' stage of the Spanish aid system (so as to focus the study on the time when it was already consolidated) and the current economic crisis (which has slowed down the immigration flows temporarily and reduced the ODA budget drastically). 28 For those variables with one-or two-year lags, the information starts from 1997 or 1996, respectively.

Main results: the role of immigration in the Spanish pattern of aid allocation
The selection of Spain's aid-partner countries (the first decision stage) has been rather 'eclectic': it has paid attention to Spanish foreign policy interests (especially immigration interests and post-colonial links), and shown a strong inertial motion and a 'gregarious' behaviour (Table 2). However, the selection neglected other factors related to the effectiveness of aid and the relative needs of the recipientsgiving priority, in fact, to the smallest countries.
The OR column shows the odds ratio of being selected as a partner country (i.e. the ratio between the probability of being selected and the probability of not being selected) when one of the explanatory variables of the model increases one unit, ceteris paribus. 29 The highest odds ratio is associated with the aid's path dependence, which imprints a character of 'persistence' in the selection process in such a way that a one per cent increase in the Spanish ODA quota allocated the previous year to a developing country multiplies by 4.1 its odds of being re-selected as a partner. This path dependence of Spanish aid is strongly linked to the inclination towards the former colonial countries, which is further reinforced by the strong specialisation of Spanish NGOs in these same countries.
The second variable with the highest odds ratio is the Spanish colonial past. The interpretation of this coefficient in the case of two-category dummy variables is relatively simple: for two developing countries with identical values in the set of explanatory variables, but with different post-colonial links, it shows the difference between their probabilities of being selected partners. According to the estimation, the odds ratio for a former colony is 3.7 times greater than for a country without this historical link, a result that stems from the fact that only 66 of the 240 observations of the colonial past included in the panel data set (20 former colonies analysed over 12 years) 30 have not participated significantly in Spanish ODA.
Third, the selection of aid partners is directly related to the share of each developing country in the aid disbursed by other bilateral and multilateral donors. Thus, a one per cent increment in the share of global aid duplicates the odds of a particular country being selected as a partner. This result confirms that Spain is influenced by a bandwagon effect, which suggests that the regions where donors share strong geostrategic interests (mainly the Middle East and the largest developing countries) are, in turn, regions of interest for Spain. 31 Ultimately, a positive coefficient of this variable reflects a lack of coordination with the rest of the donors in the definition of their 'intervention areas'which may reinforce the existence of aid darling and orphan countries.
Fourth, the selection process is significantly influenced by immigration flows. Approximately, a one per cent increase in the immigrant population from a particular country is associated with an 18 per cent increase in the probability of being selected as a Spanish aid partner. This result stems from the importance of the historical links in the selection process; thus the majority of immigrants come from former colonial countries (mainly Latin America and Morocco). This result is also due to the intensification of migration flows sent by countries without historical links with Spain, which have received increasing amounts of Spanish ODA, as in the case of China, Algeria, Senegal and Brazil (countries with significant receptions of Spanish ODA in 36 of the 48 considered observations). Moreover, the estimation reveals that Spain has not systematically considered in its selection process other criteria relative to the needs of the recipient countries, nor other aspects of aid effectivenessat least with the expected sign. Thus, the only statistically significant recipients' needs variable is the population size, with an estimated odds ratio smaller than one, which implies that a one per cent increase in a developing country's population is associated with a 33 per cent decrease in the odds of being selected as a Spanish aid partner. This reveals the existence of a small countries bias, which is the result of the attention given to countries with special historical links with Spainprimarily countries with relatively small population sizesand the lack of attention given to some of the biggest developing countries, such as India, Bangladesh, Pakistan and Nigeria.
The control of corruption variable is statistically significant, but has an opposite sign than expected: Spain seems to favour the selection of countries with the worst records of corruption. 32 As a result, the majority of developing countries with the best performance in control of corruption have received marginal amounts of Spanish ODA. Moreover, it should be noted that Spain's former colonial countries have relatively low performance in terms of corruption: on average, in the period between 1998 and 2009, the value of the control of corruption index for these countries (excluding Chile) is −0.469, which is a lower value than that of other developing countries without post-colonial links with Spain (−0.432). 33 In terms of the goodness-of-fit of the model, the χ 2 test of the overall significance rejects the hypothesis that all the variables exert a simultaneously null effect in the selection. Moreover, the model correctly classifies 92.5 per cent of the cases, offering a reasonable fit that allows us to trust the accuracy of the estimates. 34 In fact, when we drop the immigration variable, the accuracy of the model drops to 90 per centa greater decline than the one obtained when we drop the colonial variablewhich confirms the importance of immigration flows in the selection process.
The analysis of the second decision stage reveals also that the allocation of Spain's ODA-quotas has not been purely 'altruistic' (Table 3). In this sense, the distribution has not favoured especially those countries with the highest relative needs; on the contrary, the allocation has been regressive due to the Spanish 'specialisation' in middle-income countries. Moreover, immigration flows have been highly influential, as more aid is given to those countries that send immigrants to Spain. The allocation pattern also confirms both the inertial and the bandwagon effects detected in the first decision stage.
The absence of a progressive pattern of the Spanish ODA distribution is reflected in the coefficient of the life expectancy variable: a 1 per cent increase in this indicator is associated with a 0.73 per cent increase in the aid quota. This result is due to two main reasons: the traditional specialisation of Spanish NGOs in Latin America (and, consequently, in middle-income countries), where people enjoy longer life expectancy than in the rest of the developing world; 35 and a distribution biased towards those developing countries outside the Americas with the highest levels of development. For example, of the 15 countries with the highest shares of Spanish ODA, all, with the exception of Bolivia, have life expectancies greater than 68 years, including those non-Latin American countries (such as Morocco, China, Turkey, Indonesia, Bosnia and Herzegovina, and Algeria). By contrast, the countries that have received small amounts of Spanish aid sporadically are precisely those with the lowest life expectancies (mainly Sub-Saharan countries).
Moreover, immigration flows have a significant influence on the distribution of Spanish ODA. A one per cent increase in the immigrant population of a particular developing country implies an increase of about 0.05 per cent of Spanish ODA disbursed to this country. Therefore, Spain gives a strong priority to the selection of migrants-sending countries as aid partners, and eventually translates this priority into allocating them slightly greater aid quotas. Again, this result is linked to the Spanish specialisation in Latin Americathe source of the majority of its immigration flows. Especially outstanding are the cases of Peru, Honduras, Bolivia, Ecuador, Colombia, Venezuela and the Dominican Republic; all of them receive large amounts of Spanish ODA and have significant numbers of immigrants living in Spain (see Appendix B). However, this pattern is not confined to Latin America; countries like China, Morocco, Algeria and Senegal also receive large amounts of aid and send large numbers of immigrants to Spain. Furthermore, the distribution of Spanish ODA has been affected by the bandwagon effect: a 1 per cent increase in the aid received from other donors increases, on average, the Spanish aid quota by 0.23 per cent. This is despite the particular Spanish pattern of aid allocation, which is focused on former colonial, middle-income countries. In fact, this 'herding' behaviour stems from the allocations to countries outside Latin America, where Spain has coincided with the donor community in the disbursement of large amounts of aid to China, Indonesia, Serbia, Montenegro, Iraq and Mozambique. This trend coincides with a period of strong expansion of the Spanish ODA budget that has allowed an increase in the number of partner countries (at the expense of the resources' dispersion), to the point that Spain has disbursed ODA to an average of 107 developing countries per year (out of 163 possible recipient countries according to the DAC list).
The allocation of Spanish ODA has also a significant inertial movement: a one per cent increase in the previous year aid share involves an additional increase of 0.2 per cent in the current year. This result may stem from Spain's interest in maintaining (and enhancing) cooperation with its key partners, be it to minimise the administrative costs of aid, to provide stable resources that facilitate a sustainable source of resources, to emphasise those previous experiences that were particularly effective, or to avoid 'delicate' diplomatic conflicts as a consequences of retracting resources from long-running aid partners. However, importantly, many of these considerations are inconsistent with the increasing geographical fragmentation of Spanish aid that stems from an increasing number of countries receiving marginal resources, which in turn raises the management costs of aid. 36 With respect to the model's goodness-of-fit, the overall F test is flatly rejected. Moreover, the Sargan and Hansen tests for joint validity of the instrument are not rejected, as it is the case for the Arellano and Bond test for autocorrelation in the idiosyncratic disturbance term. Hence the results of these post-estimation tests support the validity of the estimations.
Also, the estimation offers reasonable adjustments, rendering a limited number of outliers. A careful examination of the outliers yields three conclusions. 37 First, there is a marked divergence between the priorities of aid loans (which are managed by the Ministry of Industry) and other bilateral ODA, which 'blurs' the distribution map of Spanish ODA. Second, most of the outliers correspond to cases of 'over-allocation' (compared to the historical pattern of distribution identified by the model), which results in an upward bias of the estimations; therefore, the regressive allocation pattern previously explained is reinforced by the number of outliers among those countries with the highest life expectancies, such as Honduras, Morocco, Turkey, Algeria, Serbia and Montenegro, Tunisia, the Philippines, and Vietnamall of them with life expectancies greater than 69.5 years. And third, it is a striking fact that most of the outlier countries have been eventually 'prioritised' by the successive Aid Plans, thus 'formalising' the deviations from the Spanish historical geographic strategy.

Conclusions
Although the 'eclecticism' of the geographical distribution of aid seems to remain the same, donors' priorities have actually evolved, 'reshaping' the aid map of the 21st century. One of the most relevant phenomena is the intensification of migration flows sent by developing countries and received by OECD countries. However, to date, the influence of immigration on geographical allocation of aid has been little studied, despite being one of the most dynamic variables of the globalisation process and an issue of great concern for developed countries. In this sense, the reception of immigrants from countries that are not historically linked to the donor changes the definition of the traditional foreign policy priorities. Thus new aid-demand factors (greater importance of the development needs of those countries that are sending migrants) and aid-supply factors (donors' own political interests as aid donors and immigrant recipient countries) are taking shape and affecting the pattern of aid giving. This new pattern is particularly interesting for OECD countries with important immigration populations, like France, the United Kingdom, Germany, the USA and Spain.
The case of Spain is particularly interesting in studying this problem, as immigration has been a phenomenon of great importance there since the 1990s. In a short time, Spain became one of the top-ten DAC donor countries and one of the main recipients of immigration in Europe. As a consequence, Spain started to disburse ODA to new 'migrants-sending countries' (mainly Sub-Saharan), which were added to the traditional list of aid partners that share historical and cultural links with Spain (Equatorial Guinea, the Philippines, Morocco and Latin American countries). Furthermore, over time, Spain has included among its aid priorities a specific sector devoted to migration and development. Spain also added some of these emigrating countries to its geographical priorities, thus 'officially' integrating them into the resource planning and evaluation cycle. However, this implies an increase in the number of prioritised countries and in the number of aid recipient countries (on average, 107 countries per year).
We study the influence of immigration in the Spanish ODA allocation by means of a two-stage decision-making model: in the first stage, Spain selects her aid-partner countries and, in the second stage, Spain decides the share of aid to allocate to each partner country. The estimation of the model rules outas expectedthe existence of a purely 'altruistic' pattern of aid allocation; on the contrary, Spain's aid map is rather 'eclectic'. As Figure 1 summarises, the allocation is not guided, in general terms, by humanitarian criteria or efficiency reasons. Instead, the distribution is determined by other foreign policy interests (post-colonial links and, more recently, immigration interests), the aid's path dependence and the bandwagon effect that stems from the insufficient geographical coordination among donors.
In particular, the main novelty of this study is the confirmation of the importance of immigration in the delimitation of the world map of Spain's ODA. Developing countries with large populations living in Spain are more likely to be aid partners (with an odds ratio of 1.18) and receive higher levels of ODA (with an elasticity coefficient of approximately 0.05 per cent). Thus, immigration flows are determinant in both decision stages (selection of aid partners and allocation of aid quotas), as it is the case of only two other variables (share of global ODA and previous Spanish ODA allocation; see Figure 1). This result relates to the importance that Spain places on historical ties, as the majority of the immigrants come from countries with 'postcolonial' links (mainly Latin America and Morocco). In addition, this result corresponds to the intensification of migration flows stemming from developing countries that are not historically linked to Spain, which have received increasing shares of Spanish ODA, as in the case of China, Algeria, Senegal and Brazil. Share of Spanish ODAt-1 Colonial past Odds ratios of being selected (first decision stage) Elasticities of aid-quotas (second decision stage) Figure 1 How much the 'odds ratio' and the 'aid quota' increase when a partner country increases its … by 1 per cent?
Note: The figure only shows those variables that are statistically significant at the 10 per cent level (see Tables 1 and 2).
Overall, the consideration of immigration flows in the geographical distribution of Spanish ODA may be appropriate if this serves as an additional indicator of the degree of aid need, thus allowing allocations to be modulated among the already prioritised partner countries. Yet, if Spain wants to progress in the definition of a truly 'strategic' aid allocation map so as to boost the impact of its interventions, first, the consideration of the immigration variable should be limited to the group of countries previously identified as priority partners, and second, the government should not 'force' the incorporation of new priorities, as this may open the Pandora's box of constant expansion of the Spanish ODA mapin itself insufficiently 'strategic'and exacerbate the problems of resources' fragmentation.
1 The main explanatory factors of Spain's immigration have been rapid economic growth between 1993 and 2008 (which required to incorporate new workers within the context of an expanding labour market), cultural and linguistic affinities with Latin America, mild climate, geographical proximity to Africa and the wage gap between Spain and the immigrating countries (Cebrián 2009). 2 In Europe, only four smaller countries (Estonia, Latvia, Cyprus and Luxembourg) have higher rates of immigrant population than Spain (Eurostat 2012). 3 See the literature reviews of Tarp et al. (1999), McGillivray (2004 and Tezanos (2008a). 4 The assumption about the 'migrant lobbyism' in the country of reception has been studied extensively.
Several studies have documented different forms of transnational linkages between immigrants and their countries of origin, including the creation of alliances with other movements across national boundaries (such as NGOs, political parties, etc.) in order to strengthen the support to their countries of origin. Three good examples are the studies of Portes (1996), Itzigsohn (2000) and Landolt (2008). 5 As Martin and Taylor (1996) first explained, growth does not necessarily reduce the countries' incentives to emigrate if there is a 'hump-shaped pattern' of migration, which implies that, in relatively poor countries, an increase in per capita income will be associated with higher emigration by allowing poor emigrants to afford the costs associated with migration. 6 The use of a 'selection threshold' follows the approach of Tarp et al. (1999). 7 The existence of decreasing marginal returns guarantees that the donor will not concentrate all its resources on one recipient: the one with the highest score in the attraction index. 8 There is not an a priori reason for the parameters of these two equations to be the same. 9 See Tezanos (2008b) for a detailed review of the studies that have analysed the Spanish pattern of aid giving. 10 See, among others, the reviews on aid effectiveness of Tarp (2010) and Tezanos (2010). 11 Policy statement by DAC aid ministers and heads of aid agencies on development cooperation in the 1990s, available at http://www.oecd.org/LongAbstract/0,2546,en_2649_34435_2755285_119814_ 1_1_1,00.html, accessed 25 July, 2012.
(Palestine and Western Sahara), countries that lack statistical information (Cuba, Kosovo, North Korea, East Timor, Myanmar and Zimbabwe) and islands and regions with fewer than 1 million inhabitants which have received very limited attention from Spanish assistance. 28 The analysis is focused on the period in which the immigrations flows were more intense, thus the estimations may be overestimated in comparison to the whole period of Spanish aid. 29 Formally, when variable x i increases one unit, ceteris paribus, the odds ratio is multiplied by a factor equal to e xi . 30 Cuba is not included in the analysis due to lack of information. 31 Such is the case of countries like Egypt, China, India and Indonesia, whichexcept for Indiahave been Spanish aid partners throughout the eleven years analysed. 32 Larrú's study (2011) has also detected Spain's tendency to give aid to countries with problems of government corruption. 33 We exclude Chile from this calculation because this country did not reach the level of significant participation in Spanish aid in any year, and because its positive results in terms of control of corruption bias the average of the group upwards. 34 Results available upon request. 35 However, Tezanos (2008c: Ch. 2) has proven that the Spanish aid allocation within the group of former colonial countries is especially progressive. 36 In fact, in the period from 1998 to 2009, on average, the aggregated aid of the 23 countries with the highest aid shares accounted for 75 per cent of Spain's ODA. Consequently, the other 83 partner countries shared between them the remaining 25 per cent of resources. 37 Result available upon request.  (1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)