The international community has been increasingly attentive to aid effectiveness since the end of the Cold War. Such concerns led to the 2005 Paris Declaration on Aid Effectiveness, a global commitment to improve the quality of aid and its impact on development. Among other threats, the Paris Declaration highlights aid fragmentation and recommends that donors coordinate their efforts in terms of comparative advantage at sector or country level. However, fragmentation has shown no signs of abating but has rather been on the rise: The average number of unique donors present in a given aid-recipient country has risen from 20 in 2004 to 30 in 2013 (Tierney et al. 2011).

Empirical evidence for fragmentation’s impact on aid effectiveness is mixed. On the one hand, studies have shown that aid fragmentation is negatively associated with bureaucratic quality and economic growth, and that it can increase corruption and hamper the effectiveness of technical assistance (Knack and Rahman 2007; Djankov et al. 2009; Annen and Kosempel 2009; Kimura et al. 2012; Kangoye 2013). On the other hand, recent work suggests that fragmentation can be beneficial when it comes to promoting child survival (Han and Koenig-Archibugi 2015) or democracy (Ziaja 2020). We posit that these mixed findings may reflect differences across aid sectors and spatial levels of implementation.

To demonstrate the conditional nature of fragmentation’s impact, we first compare results for two outcomes with widespread donor buy-in: child survival and good governance. However, as we describe in greater detail below, while donors tend to agree on the importance of achieving these outcomes (indeed, they both feature among the Sustainable Development Goals), strategies for achieving them tend to differ. Aid to the health sector increasingly focuses on achieving specific targets (Honda 2013) often using pre-specified inputs (Miller and Babiarz 2013) whereas governance aid tends to be more process-oriented.

We further examine how the level of implementation may condition results — comparing the correlates of fragmentation across and within countries. First, cross-national analysis of panel data for 152 countries from 1995 to 2014 suggests aid fragmentation can promote child survival and improve governance. However, just looking across countries has the potential to blur important subnational variation. We thus examine variation in fragmentation and outcomes within two countries (Sierra Leone and Nigeria), leveraging geo-coded data on the location of aid projects. Here, we find that donor proliferation is associated with worse health outcomes (measured in terms of child mortality), but better governance outcomes (measured in terms of budgetary stability). This is consistent with the idea that having more donors within a locality is beneficial when they are working to improve the systems through which policies are implemented, but harmful when they try to influence policy implementation directly to achieve measurable goals.

To explore potential mechanisms through which fragmentation exerts differential effects, we analyze data from a recent survey of 4,100 Nigerian civil servants. The survey records the respondents’ places of work, and thus allows us to link their responses to the number of donors present in their subnational administrative region (Nigerian states). We find that in states with more donors in the health sector, local health officials report facing increased pressure to change projects. They also report being unable to identify quality contractors for their own projects suggesting a form of internal “brain drain.” In contrast, local officials in states with more donors in government and civil society (GCS) aid face fewer barriers to policy implementation across a range of subnational public institutions. In particular, interviewees report less pressure to choose a particular contractor.

Our findings make a number of contributions to the literature on foreign aid effectiveness, and the study of development more broadly. First, our study suggests caution against making inferences based on aggregatesFootnote 1 or cross-national patterns. We also provide evidence suggesting that different aid sectors have different political economy implications, building on the work of Jones and Tarp (2016) and Gehring et al. (2017).Footnote 2 Finally, we add to emerging literature on subnational aid effectiveness (Kotsadam et al. 2018; Dreher and Lohmann 2015; Briggs Ryan 2017), which to date has not examined aid fragmentation in detail.Footnote 3

This paper proceeds as follows. “Benefits and Drawbacks of Aid Fragmentation” reviews the literature on aid fragmentation, highlighting recent empirical work. “Understanding Diverse Effects of Donor Diversity” then presents our argument, outlining conditions under which fragmentation can be expected to have a beneficial or harmful effect on the achievement of different development goals. “Empirical Strategy” describes our research design. “The Macro View: Cross-country Analysis” presents the results of our cross-country analysis, and “Zooming In: Subnational Analysis” our subnational study. “Exploring Mechanisms Through Surveys with Bureaucrats” investigates mechanisms at the level of subnational organizations, and “Conclusion” concludes.

Benefits and Drawbacks of Aid Fragmentation

The Paris Declaration includes 56 commitments to enhance aid effectiveness. One of the most well-known relates to aid harmonization, justified by the notion that, “Excessive fragmentation of aid at global, country or sector level impairs aid effectiveness. A pragmatic approach to the division of labor and burden sharing increases complementarity and can reduce transaction costs” (The Organisation for Economic Co-operation and Development 2005, 6).

This justification reflects what has emerged as near consensus regarding the negative impact of aid fragmentation, which is understood to stem in part from a set of collective action problems. That is, when many donors contribute to a given goal, responsibility for success or failure is diffused, and thus each individual donor has only a limited stake in the recipient country’s progress. Accordingly, as the number of donors increases, so do incentives for any one donor to shirk on activities that maximize overall wellbeing in favor of activities that contribute to donor-specific goals. Focusing on donor-specific goals can reduce aid effectiveness by wasting resources, for example by tying aid to the employment of donor-country contractors. Aid fragmentation can also undermine the quality of governance or slow the development of public sector capacity if it leads donors to provide aid through projects rather than general budget support, rely on expatriates instead of contributing to local knowledge by hiring local staff, and fund unsustainable investment projects (Knack and Rahman 2007). The presence of multiple donors also frequently creates undue burdens on local bureaucracies (Kanbur 2006; De Renzio and Hanlon 2008), deprives them of competent staff due to “internal brain drain” (Lemay-Hébert et al. 2020), and results in a duplication of efforts (Annen and Kosempel 2009; Makinde Olusesan et al. 2018). Moreover, Gibson Clark et al. (2005) argue that having more donors increases the recipient government’s negotiation power. As a result, donors become less demanding in selecting and supervising projects and it is easier for corrupt officials in recipient countries to expropriate resources.

There is considerable empirical support for these propositions — particularly when it comes to the negative impact of fragmentation on economic growth (Djankov et al. 2009; Kimura et al. 2012; Gehring et al. 2017). Aid fragmentation has also been shown to undermine the effectiveness of foreign technical assistance (Annen and Kosempel 2009). As for more specific effects on recipient governments, Djankov et al. (2009) find that donor fragmentation is associated with increased corruption, a result confirmed by Kangoye (2013). Gehring et al. (2017), in contrast, using a more current sample of countries, do not confirm a detrimental effect of fragmentation on bureaucratic quality detected earlier by Knack and Rahman (2007).

Some recent empirical work suggests that fragmentation can in fact be beneficial to the achievement of particular development goals. For instance, Han and Koenig-Archibugi (2015) show that countries with a moderate number of donors in the health sector fare better than those with either few or many donors when it comes to child survival. Gehring et al. (2017) detect a positive effect in the education sector. Ziaja (2020) finds that fragmentation can enhance the effectiveness of democracy aid. He argues that such aid fosters democratization by empowering local actors with resources and information. Having more donors makes this process more effective by generating a “marketplace for idea support” to help build a resilient democratic regime, rather than a top-down blueprint approach emanating from one or few donors.

While they make somewhat different theoretical arguments, all three of these studies rely on the notion that the presence of multiple donors allows for the exchange of ideas, from which recipient country governments can draw to craft locally appropriate policy solutions. What has yet to be shown is why these positive results don’t appear to transfer to other development goals, and more generally under what conditions we can expect the benefits of fragmentation to outweigh the costs. The present study begins to map a research agenda to answer such questions.

Understanding Diverse Effects of Donor Diversity

To understand when fragmentation is likely to benefit or hinder aid effectiveness, we argue that two forms of disaggregation are important. First, we argue that fragmentation is likely to exert different effects for different aid sectors — in particular, whether aid is targeted toward policy sectors that have unambiguous, measurable targets, or toward improving processes that facilitate overall policy implementation (the measurement of which is often open to interpretation). This in turn implies that the results from studies looking at a particular sector cannot necessarily be taken to apply more generally. Secondly, not only what donors do matters, but also at what scale. The level of government at which donors are presumed to be operating has implications for the appropriate level of analysis. This further implies that the results of cross-national studies may not represent valid inferences about the implications of aid fragmentation within countries.

Development Goals: Project vs. Process Orientation

While much of the aid effectiveness literature has analyzed the impact of aggregate aid, more recent contributions call our attention to differences in aid sectors, and the implications for different targets of assistance. We argue that an important consideration is whether aid is provided to further the implementation of a specific policy (through dedicated projects) or whether money is allocated with the objective of improving the institutions and processes through which various policy outcomes are achieved. Most aid sectors remain heavily project-based and thus oriented to achieving specific, measurable targets. This reflects increased attention to evaluating the effects of aid (Manning and White 2014), which is baked into international targets like the Sustainable Development Goals via an explicit global indicator framework.Footnote 4 That project-target orientation, however, may undermine sustainable, holistic policy implementation sharpening donors’ divergent self-interests (Spicer et al. 2020) and leading them to bypass recipient governments (Dietrich 2013). Even initiatives adopting a partnership approach, such as the Global Fund to Fight AIDS, Tuberculosis and Malaria, have been cited for maintaining vertical programs and introducing parallel systems and processes (Spicer et al. 2020). In Malawi, Fischer and Chasukwa (2020) show how donor-funded programming in the health system created parallel structures that undermined donors’ stated goal of “health system strengthening.” Fragmentation in project-oriented sectors may therefore lead to the proliferation of projects and multiply the burden of monitoring and evaluation.

In contrast, when aid is targeted to facilitate how policies are implemented — i.e., to improve the governance of recipient country institutions — donor diversity may be beneficial. We call this mode of delivery “process-oriented aid.” The complexity of government administration and state-society interactions make it very difficult to design a singular intervention that will suit a given recipient country. Having a multitude of donors thus increases the chances that some donors will provide aid that improves the process. In addition, the chances that donor proliferation in governance aid causes insurmountable dissent are low, as there is a consensus on the importance and principles of “good governance” within the (traditional, Western) donor community. Indeed, looking at cross-country evidence, Jones and Tarp (2016) show that in contrast to other types of aid, stable flows of governance aid display a positive association with the quality of political institutions on average. Moreover, such aid tends not to be partisan, and is often directed toward more technocratic aspects of governance, including public finance, public sector management, and decentralization (Hout 2012).

Ziaja’s (2020) results for democracy aid are pertinent here as well. He finds that the presence of more donors supports endogenous institution-building processes by creating a “marketplace for democracy support.” In such a setting, recipient country actors are more likely to find suitable support for their own, locally crafted ideas, than in situations with fewer donors.

Level of Implementation and Analysis

Beyond disaggregating the study of aid fragmentation by sector, we argue that it is important to move beyond cross-national analysis (which constitutes the majority of the literature in this field). Such a focus potentially blurs important subnational variation. The assumption that donor diversity and target outcomes develop uniformly within countries does not hold when looking at where specific donors engage. As an example, Fig. 1 shows the number of bi- and multilateral donors in the health sector for the two countries that we investigate in further detail below. Each plot’s subtitle lists the total number of unique health donors reported in the national-level dataset provided by AidData (Tierney et al. 2011). The maps show the number of unique health donors present within each subnational unit as reported by the respective Aid Information Management Systems (AIMS).Footnote 5 Sierra Leone records 19 donors in the health sector on the national level, and from eight to 18 in each district, resulting in a substantial standard deviation of 3.0. In Nigeria, a much larger country, the density is much lower and more uniform, ranging from two to eight per state with a standard deviation of 1.4. In contrast, the number of donors recorded on the national level amounts to a much higher number of 27.Footnote 6 Any analysis of aid fragmentation conducted at only one level would thus be vulnerable to the “modifiable areal unit problem” (MAUP). That is, fragmentation may manifest in very different ways — within and across countries — depending on the level of spatial aggregation, which can in turn affect inferences drawn from using different measures (Lee and Rogers 2019).

Fig. 1
figure 1

Number of health donors by subnational administrative units

In substantive theoretical terms, we also expect differences in how aid fragmentation plays out across different levels of government. National administrations tend to be staffed with more senior personnel than subnational entities. In addition, most political power in aid-recipient countries resides at the national level, even if some degree of decentralization may exist (Bossert 1998; Olowu and Wunsch 2004; Dickovick 2005). The key players involved in aid negotiations typically include cabinet members and other powerful officials used to operating on the international stage (Gibson Clark et al. 2005). Such actors also have the power to introduce or influence legislation, which may be conducive to managing the detrimental effects of fragmentation. National administrations are thus arguably in a better position to take more ownership of the aid implementation process, compared to their lower-level counterparts.

At the subnational level, various detrimental mechanisms of project proliferation have been shown to materialize, hindering policy implementation.Footnote 7 For example, a study of global health initiatives found that donors frequently bypass local governance structures, leading to service duplication (Spicer et al. 2010). More donors with more projects hire more qualified local staff, draining recipient institutions of their best personnel (Lemay-Hébert et al. 2020). The desire of aid managers to guarantee the success of their projects decreases concerns for collateral damage and opens doors for corruption. Add to that the possibility of having seasoned aid project managers with strong and mutually incompatible views on how to implement particular policies, a local vicious circle of project competition may emerge, undermining aid effectiveness in fragmented settings.

At either level of government, however, we might expect the positive, “marketplace for ideas” aspect of aid fragmentation to dominate — though this seems more likely for some sectors than others, as discussed above. More generally, expecting results found on the national level to hold when we look at the subnational level may be misguided depending on how donors are distributed within a country.

Empirical Strategy

We explore the impact of these different dynamics of aid fragmentation — level of implementation and process-vs.-project orientation — using data at three different levels of aggregation from two aid sectors. Specifically, we examine the consequences of aid fragmentation at the national level, at the level of subnational administrative divisions, and at the level of implementing organizations in recipient countries. Our multi-level approach has a combined methodological and conceptual purpose. Methodologically, using several spatial levels decreases vulnerability to the MAUP. We can see whether results differ at different spatial resolutions. Conceptually, we already expect such differences, as argued above: Donor crowding at the national level may affect implementation in different ways than it does on the local level. Finally, we analyze survey responses of Nigerian civil servants across local (state) organizations, which sheds light on the mechanisms driving the varied impacts of fragmented aid.

At the same time, we look at two aid sectors — one representing a prototypical project-oriented type of aid and the other, a more explicitly process-oriented category. Our choice of sectors and outcomes analyzed is further driven by data availability, as well as a desire to examine aid targeting outcomes that enjoy widespread support in the donor community. As such, we examine health aid, which ideally contributes to reducing child mortality, and government and civil society (GCS) aid, also known as “governance” aid. Table 1 summarizes our sectoral setup.

Table 1 Target type, aid sector and code, and target outcome

Our main measure of aid fragmentation is the number of donors present within a unit of observation. To count donors and measure aid by sector, we draw on data generated by AidData (Tierney et al. 2011).Footnote 8 At the national level, we count donors to be present in an aid sector only if they commit to spending at least 100,000 USD per year. We expect that this is approximately the level at which donor engagement becomes notable, e.g., via the presence of an expatriate expert. At the subnational level, we abstain from setting such a threshold. Donor personnel could travel regularly from neighboring regions even where low amounts of money are spent. We include all bilateral OECD donors recorded in the data, as well as multilateral and private donors.Footnote 9 The universe of cases we study comprises all countries eligible to receive aid according to the OECD-DAC’s official list of recipients.

The estimation methods we employ for assessing the relationships between explanatory and explained variables are designed to strike a balance between data quality, data availability, and inferential rigor. In the absence of a true or a natural experiment, we cannot establish causality. Insights about temporal patterns in aid allocation and aid effectiveness, however, allow us to design correlational analyses that approximate what we expect to be the underlying data-generating processes (Clemens Michael et al. 2012). We thus draw on bivariate correlations with substantial time lags, estimate fixed-effects regressions, and analyze a range of complementary indicators from the civil servants survey.

The Macro View: Cross-country Analysis

We begin by looking across countries to compare the consequences of fragmentation in the two aid sectors of concern. As most outcomes considered here are rather slowly moving, and since we want to guard our analysis from the most obvious channels of self-selection and reverse causation, we measure outcomes in terms of 10-year averages and lag the number of donors by one 10-year period. Specifically, we observe the total number of unique donors in a particular sector in the period 1995–2004, and change in the average level of the respective development outcome from that same 10-year period to the following 10-year period, i.e., change from 1995–2004 to 2005–2014. Although this does not provide a causal estimate, this approach nonetheless provides us with a first impression as to whether there is a sustained association between the number of donors and outcomes.

Child mortality — our target for outcome-focused aid — is measured as the probability per 1,000 that a newborn baby will die before reaching age five (Sharrow et al. 2022). It has declined in virtually all countries in our sample from the period 1995–2004 to the period 2005–2014, as shown in the first panel in Fig. 2. Change is also strongly associated with the number of unique donors that had worked in health aid in the period 1995–2004. We see that a greater number of donors is significantly correlated (p < 0.01) with greater reductions. Child mortality is expected to be ten percentage points lower when ten donors or more are present.

Fig. 2
figure 2

Relationships between the number of donors and changes in outcomes at the national level

The number of donors in GCS aid — our process-oriented aid sector — and its relationship with government effectiveness (as measured by the Worldwide Governance Indicators) are shown in the second panel. Countries with many GCS donors exhibit considerable variation: South Africa and Zimbabwe deteriorated while China, Ethiopia, and Rwanda improved. On average, having ten more donors is correlated (p < 0.10) with an increase of about 0.1 on the government effectiveness scale (which ranges from − 2.2 to 2.0 across the two 10-year periods considered here).

As one might expect the number of donors to have a more immediate effect, we repeat the exercise employing the contemporaneous number of donors (2005–2014). Direction and size of the relationship remain very similar and significant for both pairs of variables (see Figure A3 in the online appendix).

In sum, both long-term correlations reach a significance level we consider suitable for the number of observations in our sample, and both imply that having more donors in a given sector correlates with better outcomes, both in the long term and contemporaneously. This is at odds with the received wisdom about the negative consequences of aid fragmentation but in line with the more recent studies we cite above.

These bivariate snapshots are vulnerable to omitted variable bias as a number of country-specific factors may affect both the number of donors and the development outcomes they aim to address. Some of these are slow to change, and thus can be accurately captured through country fixed effects. Some are global shocks that can be captured with year fixed effects. Others show movement within countries in ways that could affect our outcomes variables of interest. We thus turn to a multivariate analysis that incorporates fixed effects and the relevant control variables.

The first, and perhaps most obvious important factor to control for is the overall amount of aid being directed towards a given goal in each country, being related both to the number of donors and outcomes. Relatedly, income per capita is a major determinant of donor involvement and has been identified as a critical variable in explaining variation in key summary measures of population health, such as infant mortality, under-5 mortality, and life expectancy at birth (Croke 2012). Conflict prevalence affects donor presence and has also been cited as a historical driver of poor child health outcomes (Croke 2012), and may affect the achievement of other development goals. We thus include a dummy for ongoing civil war as defined by the 25-battle-deaths threshold of the Uppsala Conflict Data Programme (Themnér and Wallensteen 2012). Finally, larger recipient countries tend to have more donors than smaller countries. But population size is unlikely to directly affect health or governance outcomes. We thus do not control for population.Footnote 10 Table A1 in the online appendix provides summary statistics for all variables included this dataset.

Employing these control variables leaves us with data for 152 countries over the period 1995–2014. The temporal restriction is due to the availability of sector-specific aid data: before the early 2000s, sectoral decomposition of aid is fairly unreliable; before 1995, it is unacceptably incomplete. Instead of relying on two 10-year cross-sections, as for the scatterplots above, we now want to learn from temporal variation and thus consider shorter time periods. Given that our data is based on commitments, however, we need to allow for a certain lag, as aid commitments are frequently realized several years after the decision was recorded. This also makes commitments more volatile than disbursement data. Moreover, the effects we want to observe are unlikely to materialize after just 1 year. We thus opt for 3-year periods as our temporal unit of observation. Combined with a one-period lag, we are confident that at least the most pressing endogeneity concerns are addressed by our specification (cp. Clemens Michael et al. 2012).

Table 2 presents the regression results. We present results without and with controlling for aid per capita (panels A and B, respectively). Results are presented in this manner because controlling for aid when examining the effects of the number of donors in the same sector causes a statistical problem (Ziaja 2020, 441). The amount of aid is an intermediate outcome on the path from the number of donors to the target outcome: A donor first decides to start relations with a recipient country (contributing to the number of donors) and then to commit a certain amount of aid (contributing to the overall amount). Intermediate outcomes constitute “bad controls” that introduce additional selection bias into the comparison of recipient country democracy conditional on the number of donors (Angrist Joshua and Pischke 2009, 64–68). At the same time, controlling for aid is essential from a substantive point of view, as argued above. A solution to this dilemma is focusing on the interaction between the amount of aid and number of donors as main explanatory variable. We do not even need to include the constitutive terms of this interaction: The number of donors is only and always zero when the amount of aid is zero, by definition. This relieves us of the duty to include constituting terms, and of the necessity to interpret the interaction results visually. The mere significance of the interaction’s coefficient provides substantial evidence of a relationship pattern not induced by chance.Footnote 11 Panel C implements this specification.

Table 2 Regression results: the number of donors and outcomes on the country level

Panel A shows models with the number of donors in the respective sector as the only explanatory variable. The amount of aid is not yet included. Fragmentation in the provision of both health and governance aid is significantly related to their respective outcomes and has the desired effect (more donors, better outcomes). Panel B repeats the same specification controlling for amount of aid. Aid amounts are measured in per capita terms, as our outcome child mortality is an individual-level trait. The same level of absolute aid would mean very different things for a country with a small population than for a much larger country. Amounts are also logged to account for decreasing marginal returns. Results from Panel A are largely confirmed: the number of donors providing health and GCS aid is significantly related to better outcomes. Aid per capita is only significant in the model using health donors as explanatory variable. Finally, Panel C reports the estimates of the interaction of the number of donors in the respective sector and the logged amount of aid per capita provided in this sector. Both health and GCS donors and aid maintain their significance.Footnote 12

These findings confirm the picture obtained from the scatterplots above. Our results on the national level thus complement the findings of Han and Koenig-Archibugi (2015) and Ziaja (2020): health aid effectiveness appears to benefit from donor diversity, as does GCS aid. In the next section we proceed to investigate whether these associations hold when we look at variation within countries.

Zooming In: Subnational Analysis

As noted above, cross-national analysis may not capture the true consequences of aid fragmentation on the ground. We thus turn to subnational aid allocation patterns in two countries in sub-Saharan Africa. We focus on this region as it is home to the largest recipients of net official development assistance, and is also where many prior studies on the pernicious effects of aid fragmentation have focused (O’Connell Stephen and Soludo Charles 2001; Bräutigam Deborah and Knack 2004; Kimura et al. 2012). We examine two countries representing different levels of aid-dependency that have sufficient subnational data on aid flows and appropriate outcome indicators: Sierra Leone and Nigeria.

Sierra Leone is small in terms of geography and population, low-income, and aid-dependent. Net official development assistance (ODA) accounted for 13% of GNI in 2018, or USD $66 per person. As of 2013, Sierra Leone registered 42 bi- and multilateral donors, including 19 in health and 27 in GCS aid.Footnote 13 The country’s post-conflict status presents particular opportunities and challenges for donor diversity. Following the 1991–2002 civil war, the country’s peace process is seen as having benefited greatly from an “eclectic mix” of donors, whose substantial contributions addressed critical needs such as the resettlement of refugees, the disarmament and reintegration of combatants, and the rebuilding of state and civil institutions (Kanyako 2016). As in many other post-conflict settings, both the government of Sierra Leone and its donors recognized the need to rapidly rebuild the health system and increase health service provision, both as a goal in itself as well as a catalyst for peace building. However, many of the policies that emerged during this time were seen as externally driven, lacking local ownership necessary for their effective implementation (Bertone Maria et al. 2014).

Nigeria, in contrast, is large in territory and population. Nigeria also records large income from oil exports, dwarfing the substantial amounts of absolute aid that it receives to just under 1% of GNI and USD $17 per person in 2018. As of 2013, Nigeria registered 44 bi- and multilateral donors, including 27 in health and 25 in GCS aid. Notably, various interventions have been put in place in recent years to address donor coordination — particularly in the health sector. These include developing a National Strategic Health Development Plan (NSHDP), which established a Health Development Partners Forum as a mechanism for donors and the government to coordinate interventions. The NSHDP further identified the need for coordinating funding mechanisms using a sector wide approach. The country has also undergone a rationalization exercise aimed at allocating different donors to different geographic areas (Makinde Olusesan et al. 2018). However, recent studies characterize the harmonization of aid as “dismal,” finding that Nigeria’s donors fail to align with the country’s reporting and monitoring systems (Chiegil 2017) and that a duplication of effort persists (Makinde Olusesan et al. 2018).

In considering Sierra Leone and Nigeria, our study spans the range of variation in aid fragmentation on the continent. In one case, aid dependence is high, suggesting donors play an important role in governance and service delivery. In the other, donors’ overall presence and expected influence is much more limited. Thus we can think of the two countries as representing one most-likely and one least-likely case for aid fragmentation to affect policy or processes. As for choosing the level of spatial disaggregation that is most appropriate for our analysis, we turn to the lowest level with sufficient variation in donor presence. For Sierra Leone, we examine variation across secondary administrative divisions, which comprised 14 districts during the period studied in our analysis.Footnote 14 Zooming in further is not possible, as only a fraction of tertiary divisions (chiefdoms) report aid activity. For Nigeria, we base our analysis on the variation between the 36 states and the Federal Capital Territory (FCT).

Given data limitations, we do not use the same time periods as in the national-level scatterplots above. We instead restrict ourselves to the time period for which the data seems most reliable. This levels out fluctuations that may be due to different reporting patterns between donors rather than the actual engagement we want to trace. We thus count all donors that were present from 2004 to 2013.Footnote 15

We measure under-5 mortality across subnational units with new fine-grained data (Golding et al. 2017). This data is available at high levels of resolution and can thus be aggregated to the respective administrative-unit levels and merged with the subnational aid data. Figure 3 shows how trends in under-5 mortality relate to the number of health donors present in subnational administrative divisions in both countries. As we cannot fit another 10-year period after 2013, we opt to measure changes in the explained variable in 5-year averages. Change in under-5 mortality is thus calculated as the average rate from 2013–2017 minus the average rate between 2008–2012, expressed as percentage of the average rate between 2008–2012. Note that for all subnational units in both countries, child mortality has decreased by between 5 and 30% (see the range covered by both y-axes). The x-axis reports the number of health donors. For both Sierra Leone and Nigeria, more donors correlate significantlyFootnote 16 with slower decreases in child mortality. This stands in contrast to the cross-national results reported above: whereas countries with more health donors saw faster declines in child mortality, subnational units within both Sierra Leone and Nigeria saw slower declines.

Fig. 3
figure 3

Health donor proliferation and change in under-5 mortality

Lacking a common indicator to assess how aid fragmentation correlates with governance at the subnational level, we draw on two country-specific data sources. For Sierra Leone, we extracted data from the yearly report of the Auditor General, which provides information on own-source revenues of districts. The report distinguishes amounts that were budgeted from those that were actually generated.Footnote 17 We use the gap in uncollected revenue to measure the performance of districts in planning and collecting revenue. As we aim to observe change in the explained variable, we calculate differences between 2013 and 2018 performance. For Nigeria, we identified an index of financial sustainability developed by the Nigerian NGO BudgIT, which aggregates information on state expenditures, revenues, and debts into an index.Footnote 18 The data is available for the years 2015 and 2018, from which we calculated the differences to assess change.Footnote 19

Figure 4 shows a pattern in line with a positive impact of subnational fragmentation in governance aid for both countries. Districts reduced uncollected revenue in Sierra Leone as the number of donors increased. Likewise, the financial sustainability index improved in Nigerian states in which more donors were present. These relationships, however, are not statistically significant. Moreover, the downward slope for Sierra Leone is largely driven by the Western Urban area which includes the capital Freetown, where a large number of GCS donors and a substantial drop in uncollected revenue were recorded. The positive correlation of donor proliferation with governance outcomes at the national level does not receive additional support from the subnational analysis. Unlike for health, however, results are merely insignificant and still point in the same direction.

Fig. 4
figure 4

Government and civil society (GCS) donor proliferation and change in fiscal performance

As with our cross-country exploration, we expand our bivariate to a multivariate analysis to guard us from omitted variable bias. In particular, the governance indicators we employ are sensitive to economic shocks, as is aid. Controlling for per capita income should reduce this potential bias. Like the cross-national analysis, we also control for conflict, and in the case of governance aid population of the subnational unit. In addition to these core control variables, subnational analyses open the opportunity for distinguishing between urban and rural locations. Donors may be more likely to gather in urban settings, and it is usually easier to implement health and governance measures in densely populated areas. We thus add the degree of urbanization as an additional control variable (Smits and Permanyer 2019). Data limitations in the geo-coded subnational aid datasets do not allow us, however, to employ variation over time. Our analysis thus constitutes a cross-section over the years 2003–2014. Similar to the preceding cross-country analysis, moving from a bivariate to a multivariate analysis does not change the results substantially: Fragmentation in health aid remains negatively and significantly correlated with child survival, whereas fragmentation in GCS aid remains insignificant for governance outcomes. The results tables and a discussion can be found in Section “Empirical Strategy” of the online appendix.

In sum, the results of the subnational analysis clearly differ from what we find in our cross-national study. This implies that getting the spatial level of analysis right matters for studies of aid fragmentation. Neither the national nor the subnational level of analysis is suitable to answer all questions, and researchers should consider which level is appropriate for their implied data-generating processes.

Exploring Mechanisms Through Surveys with Bureaucrats

In order to better understand the mechanisms underlying the relationships between aid fragmentation and development outcomes we observe above, we leverage data from a survey of 4,100 Nigerian civil servants conducted in 2010. The survey focused on the management practices of 63 organizations of the Federal Civil Service in Nigeria, including central ministries and regional development authorities.Footnote 20 The survey sheds light on how civil servants perceive their job situation and what kind of obstacles hamper policy implementation. While the effects of aid and fragmentation were not surveyed explicitly, information on the location of respondents’ organizations and the distinction of organization types allows us to investigate the relationship between sector-specific aid fragmentation in different Nigerian states and perceptions of civil servants in potentially affected organizations in those sectors.

We select twelve questions from the survey that represent concerns discussed in the fragmentation literature. They proxy issues such as country ownership, alignment between recipient and donor priorities, aid effectiveness, and corruption. For example, civil servants are asked whether they have control over their organizations, whether they have access to sufficient funds and qualified personnel, and whether they were pressured to change specifics of their own projects, such as location, project design, or contractors. All responses are coded to represent the share of respondents providing undesirable answers, such as loss of control, dissatisfaction, or being pressured to modify a project.

As in the preceding subnational analysis, we examine relationships at the level of Nigerian states. We plot the number of donors present in each state against the average responses provided by members of organizations based in these states. We first consider state-level fragmentation in health aid, and look how this relates to responses from employees of 16 federal medical centers spread over 15 states. In contrast to our preceding analysis that considered aid data until 2013, we drop donors that entered the country after 2010, the year in which the civil servants were interviewed. Between 2004 and 2010, the 15 states had between three and eight unique health donors. Figure 5 shows that across almost all twelve questions, undesirable outcomes increase with the number of health donors present in the state. Only the perception of not having control over their own organization (top left panel) and the necessity to use own funds decreases (bottom right panel).

Fig. 5
figure 5

Donor proliferation in health aid and civil servant satisfaction in federal medical centers in Nigeria

Only three of the 12 indicators relate to the number of donors at a statistically significant level. They all represent detrimental effects, and they directly link back to the fragmentation literature. The average share of civil servants reporting a lack of quality contractors or consultants tendering for projects increases from below 20% when three donors are present to 30% when eight donors are present (panel in row 2, column 2). This pattern is consistent with the “internal brain drain” hypothesis: donors hiring away the most talented local staff. Pressure to change the location of projects increases in similar quantities (row 3, column 2). Such pressure in states where more donors are present is consistent with the expectation that donors will prioritize achieving outcomes for their projects over local ownership. Feelings that civil servants have no control over their organization are lower where more donors are present, which is more difficult to reconcile with our expectations.

To assess the impact of fragmentation in governance aid, we consider all 95 organizations surveyed. Governance aid not only targets core political institutions such as parliament and the judiciary, it also aims to improve administrative procedures between all institutions that are part of the government or owned by it.Footnote 21 Figure 6 shows the relationship between the number of governance aid donors and civil servant responses across all organizations surveyed. The results differ fundamentally from the health sector. Eight of twelve questions yield lower shares of negative answers when more governance donors are present (panels 5 to 12, counting from the top left row-wise). Five of these relationships are statistically significant (row 3 and the first two panels of row 4). They again comprise specifications of project implementation — pressure to change project location or project design, pressure to select specific contractors, and pressure to divert of funds. All of these issues are notably less frequent in organizations located in states that host many governance aid donors. Moreover, respondents report “others breaking rules” less frequently than in settings with fewer donors. These patterns are compatible with our expected beneficial impact of fragmentation in governance aid. We ascribe this correlation to governance aid being focused on improving the process by which outcomes are achieved, rather than achieving targets associated with particular projects. Having multiple donors providing governance aid increases the options on offer and thus the chances of improving processes.

Fig. 6
figure 6

Donor proliferation in governance aid and civil servant satisfaction in Nigeria

To gain further confidence in the patterns presented in the bivariate correlations, we introduce state-level control variables (aid amounts, wealth, urbanization, and violence) in a multivariate regression framework. As described in further detail in Section “The Macro View: Cross-country Analysis” of the online appendix, we find that the correlations between the number of health donors with perceptions that there are “No quality contractors,” having “No control over organization,” and being “Pressured to change location” remain significant when adding most control variables. For GCS aid, we find that correlations remain largely significant for feelings that “Others broke rules”, being “Pressured to choose contractor,” and claims to be “Dissatisfied with job.”


In recent years, the received wisdom that aid fragmentation is harmful has been challenged, with scholars increasingly emphasizing conditionally beneficial effects. But our knowledge about the impact of aid fragmentation is still piecemeal. In this study, we take a step towards generating a more comprehensive understanding of the dynamics of aid fragmentation. The first key takeaway of our study is that both the level of analysis and aid sector under consideration can yield distinct results. As such, inferences drawn from one level or sector should not be taken as evidence of more general phenomena. Our study further suggests conditions under which fragmentation is likely to help or hinder aid effectiveness, though more and better data and creative estimation strategies are needed to generate causal inferences.

Our analysis at the national level suggests that more fragmented aid may be more effective. Looking within countries, the picture changes — particularly when we look at the level of subnational implementing organizations. Our results here are in line with expectations that project-oriented aid may be less effective than aid supporting policy implementation. Given the incentives facing donors on the ground, attributable and quantifiable project success may trump concern for overall benefits (Knack and Rahman 2007, 178). Project managers are not judged for how the recipient country fares, but whether they achieve the deliverables of their terms of reference (Nunberg and Taliercio 2012; Isenman and Shakow 2010). Having multiple implementers in the same sector in the same location blurs responsibility for negative externalities. This mechanism applies particularly to project-oriented aid.

Studying the perceptions of Nigerian civil servants allows us to further pursue this line of inquiry. For organizations in the health sector, we note patterns in line with various detrimental dynamics suggested by the aid fragmentation literature. Respondents in states where many health donors are present report pressures to adapt project locations and design. They also report a lack of quality contractors, suggesting donors hiring the most qualified local staff and leaving recipient institutions stripped of their best employees (Knack and Rahman 2007, 179). In contrast, respondents in localities with high levels of fragmentation in governance aid report mostly desirable outcomes, including less interference with various project specifications. This accords with our expectation that diversity can be beneficial for finding sustainable solutions in complex environments. A “marketplace” for reform ideas may offer greater chances of providing solutions that fit recipient needs (Ziaja 2020, 436). As there is less pressure on “getting it done” than in project-oriented aid, competition between projects may not hamper overall aid quality as it can where quality assessments are linked more closely to deliverables. Moreover, sectors without clear development goals in complex, fragmented settings favor making best use of the good judgement of field agents (cp. Honig 2018).

The rise of emerging donors such as China and India is further increasing complexity and thus presents new questions for the study of aid fragmentation. As scholars begin to tackle these questions, we urge them to take a disaggregated view — considering how both the level of analysis and target of aid may condition results.