The Impact of Decentralised Development Co‑operation

Annex D. Empirical specifications and regression results

Empirical specifications

Specification for examining cross-border decentralised development co-operation official development assistance (DDC ODA) targeting (Equation 1)

Before estimating the impact of cross-border DDC ODA on development outcomes, the analysis tests whether cross-border DDC ODA is systematically directed towards poorer regions. Because ODA is typically expected to be directed towards poorer places, the empirical approach first explores the strength of this relationship or whether it reflects differences across countries or changes over time. By comparing results across increasingly stringent models – from pooled estimates to specifications with country and year fixed effects1 – the analysis tests whether targeting holds within countries and over time. Country fixed effects are included to account for national-level factors that may jointly influence both the allocation of cross-border DDC ODA and regional income trajectories. Expressing both variables of interest in logarithms allows the coefficient to be interpreted as an elasticity, capturing the percentage change in ODA associated with differences in regional income.

Empirical specification for cross-border DDC ODA targeting:

Equation 1

Specification for assessing United Nations (UN) Sustainable Development Goal (SDG) outcomes in recipient regions and cities (Equations 2-4)

Recipient regions

The empirical analysis for regions explores whether places that receive more cross-border DDC ODA show greater improvements in development outcomes in subsequent years, while accounting for initial levels of income per capita as poorer places tend to receive more cross-border DDC ODA. To assess this relationship, the analysis focuses on changes in outcomes over time rather than on their value at any single point, comparing each region to itself over time. This approach removes the influence of fixed territorial characteristics – such as geography or long-standing institutional factors – that do not change much over time but may affect both aid allocation and development outcomes. It does not, however, capture the extent to which such time-invariant characteristics may shape how effectively ODA translates into outcomes. Cross-border DDC ODA (in USD per 100 000 people for the recipient region specifications) is also lagged by one year (set one year prior to the outcome; for example, 2018 disbursements are linked to 2019 outcomes) to account for the fact that development assistance typically requires time before its effects become observable.2 The model also includes initial income levels, reflecting that poorer regions may both receive more cross-border DDC ODA and experience faster improvements due to conditional convergence dynamics,3 as well as controlling for population size, given that larger regions may attract and absorb disproportionally higher volumes of DDC ODA and exhibit distinct development trajectories.4 As in the previous specification, country fixed effects are included to account for the influence of time-invariant national-level factors.

The empirical specification links changes in regional SDG-specific outcomes to SDG-specific cross-border DDC ODA disbursements while controlling for key structural factors. For each SDG outcome tested, the analysis restricts the sample to the most relevant type of cross-border DDC ODA based on the SDG or thematic focus of projects, poverty-related disbursements, for example, when analysing income outcomes (log of estimated income per capita, in USD 2017 purchasing power parity [PPP]). Since OECD Creditor Reporting System (CRS) projects are often tagged with multiple SDG labels and sector codes, the thematic matching captures a degree of cross-sectoral linkage. For example, a project coded under poverty, food security and health would still be included in the specification for income growth. In addition, DDC projects labelled as emergency response are excluded when computing the cross-border DDC ODA variable. This restriction is applied to reduce concerns of reverse causality, as emergency-related ODA may be allocated to territories experiencing crises, shocks or sudden deteriorations in outcomes. Including these projects could therefore make it more difficult to distinguish whether cross-border DDC ODA is associated with improvements in local outcomes or whether higher ODA reflects a response to adverse short-term local conditions. This restriction also applies to the following specifications linking SDG-specific ODA to SDG outcomes.

Empirical specification for SDG outcomes in recipient regions, year-to-year differences:

Equation 2

The analysis focuses on SDG outcomes for which cross-border DDC ODA flows have a plausible thematic link to the indicator being assessed. SDG outcomes can only be meaningfully assessed where the cross‑border DDC ODA flows considered have a plausible thematic link with the indicator under analysis. The empirical specifications therefore match cross-border DDC ODA to outcome domains according to the SDG focus of the intervention, for example by linking SDG 1-related disbursements to income indicators. This ensures that the estimated relationships are based on interventions that could reasonably affect the outcome measured, rather than on the full set of DDC projects with heterogeneous objectives and mechanisms.

Beyond the analytical relevance, the selection of outcome indicators for regions is largely determined by data availability constraints. CRS data show that a large share of DDC projects is concentrated in development sectors such as health, education and multi-sector activities, including urban and regional development, while the OECD surveys on the impact of DDC (hereafter “DDC impact surveys”) indicate a strong focus on both social and environmental objectives. At the same time, data availability – particularly for regions in ODA-eligible countries – limits the set of comparable indicators that can be used in the analysis. Building on the Global Data Lab Subnational Human Development Index, the analysis focuses on income indicators, which offer the most consistent spatial and temporal coverage across countries.

A complementary long-difference specification is also tested for income outcomes to reduce sensitivity to short-term volatility in annual ODA disbursements and income measures. In this simplified model, the dependent variable is the change in log estimated gross national income (GNI) per capita over the longest period available for each recipient region within 2013-2023, while the main explanatory variable is the cumulative amount of SDG 1-related DDC ODA per 1 000 inhabitants received over the same period. This approach smoothes year-to-year fluctuations and allows more time for income outcomes to adjust to DDC engagement. The trade-off is that, because the specification measures cumulative DDC ODA and income changes over the same period, it is likely more exposed to reverse causality than specifications using lagged disbursements (Equation 3).

Empirical specification for SDG outcomes in recipient regions, long-differences:

Equation 3

Recipient cities

The empirical analysis for cities examines whether faster growth in DDC ODA is associated with improvements in environmental and climate-related outcomes. Because city-level indicators are not available every year, the analysis measures the average annual rate of change in outcomes across years where data are available within the 2013-2023 period. This approach allows the estimation to capture changes in environmental and climate-related indicators such as green space availability (share of green area in built-up area, in percentages), and carbon dioxide (CO₂) emissions from transport (in tonnes per ten people). While transport emissions are available only for three years (2015, 2020 and 2022), green areas are available only for two years (2015 and 2020), which limits their samples. As with the regional analysis, DDC ODA (in USD per 10 000 people for the recipient cities specifications) is lagged by one year to reflect the time required for development assistance to translate into observable changes on the ground.

The model relates changes in environmental outcomes to the growth of DDC ODA while accounting for initial conditions and national contexts. As the analysis focuses on indicators related to air quality, green space and emissions, the estimation restricts DDC ODA to projects aligned with SDG 11 “Sustainable cities and communities” and SDG 13 “Climate action”. The specification controls for baseline outcome levels to capture conditional convergence dynamics, as cities with weaker initial conditions may improve faster over time, as well as for population size, given that larger cities may attract disproportionate higher levels of ODA due to greater administrative capacity and project scale, and exhibit distinct environmental dynamics. Country fixed effects are also included to account for national institutional and policy factors that may jointly influence both DDC ODA allocation and environmental performance (Equation 4).

Empirical specification for SDG outcomes in recipient cities:

Equation 4

The focus on environmental and climate-related outcomes at the city level reflects both the thematic orientation of DDC activities and current data availability. DDC and city-to-city partnerships frequently address urban sustainability challenges – including climate adaptation, disaster risk reduction, environmental protection and circular economy – aligned with SDGs 11, 12 and 13, as confirmed by CRS data and the OECD DDC impact surveys. While cities also engage in DDC activities targeting dimensions such as education, social inclusion and gender equality, the availability of harmonised indicators at the city level remains limited for these outcomes. By contrast, the Global Human Settlement Urban Centre Database (GHS-UCDB) provides globally comparable environmental data based on satellite observations and modelling, including for non-OECD contexts. This makes the environmental dimension both analytically feasible and policy-relevant for assessing DDC impacts in urban areas.

Specification for assessing local governance outcomes in provider regions (Equation 5)

The empirical specification assesses whether higher DDC ODA per capita among provider regions and stronger governance are associated. Governance outcomes are measured using indicators from the European Quality of Government Index (EQI), including overall quality, impartiality, corruption and the quality of public services, while DDC engagement is measured as log DDC ODA per capita. The specification controls for income levels and population size in the baseline year, as regions with different economic conditions and scales may both engage differently in DDC ODA (in USD per capita) and display different governance profiles. Country fixed effects are included to account for national institutional and policy environments, ensuring that the estimated relationship reflects within-country variations across regions within the same provider country rather than broad differences between countries. The results provide statistical associations rather than causal effects, as better-governed regions may be more likely to engage in DDC in the first place (Equation 5).

Empirical specification for local governance outcomes in provider regions:

Equation 5

The focus on governance for provider regions reflects both conceptual considerations and evidence from the DDC impact surveys. Unlike recipient territories, where DDC is expected to directly influence a range of social, environmental and economic outcomes, the benefits for providers are more likely to operate through indirect channels, including peer-to-peer learning, knowledge exchange, public-sector innovation and multi-stakeholder engagement, with effects likely to appear first in local governance outcomes before potentially translating into benefits in other areas. These channels are closely linked to the importance of in-kind contributions in DDC, which are not fully captured by financial flows but are central to the mutual benefits reported by provider territories. Survey evidence supports this distinction, as provider regions consistently report higher benefits in areas such as civic engagement, participation in decision making and multi-level governance, and – on top of SDGs 11 and 17 which are inherent to DDC – on SDG 16 related to strong institutions. The analysis therefore centres on indicators derived from the EQI, which capture key dimensions such as impartiality, corruption and the quality of public services. These indicators serve as proxies for the types of institutional improvements reported by provider regions.

The regression analysis for provider regions uses the EQI and its components as the main governance indicators due to data availability and temporal coverage. Initial explorations using both the EQI and SDG 16 indicators from the OECD territorial SDG framework suggest some associations between DDC engagement and governance outcomes. However, SDG 16 indicators are currently available for only two time points and for a limited set of regions. By contrast, the EQI dataset provides multiple observations over the 2013-2023 period, enabling a more robust analysis of governance performance. For this reason, the regression analysis focuses on EQI indicators, while SDG 16 measures are used as complementary evidence (see Annex E).

Regression results

The results presented in this section are based on the subset of regions and cities for which both localised DDC flows and comparable outcome indicators are available. As data coverage varies across thematic dimensions, side of the partnership and territorial levels, the sample size differs across specifications. Details on the number of units, countries and observations included in each model are reported in the corresponding regression tables. Where sample sizes are small, associations should be interpreted with caution.

Regression results for recipient regions and cities

Does DDC ODA go to poorer regions?

DDC ODA is systematically targeted towards poorer regions within countries. When comparing regions within the same country, a clear relationship emerges, with poorer regions receiving more DDC ODA per capita. Quantitatively, a 1% higher level of GNI per capita is associated with around 0.57-0.64% lower DDC ODA received across regions within the same country, indicating that resources are systematically and strongly directed towards regions with greater development needs. This pattern becomes visible only when comparisons are made within countries, as cross-country differences in allocation practices obscure the relationship between regional income and aid flows in aggregate comparisons (Table A D.1).

Table A D.1. Regression results: do poorer regions receive more aid?

Notes: Based on region-years with positive ODA disbursements. Standard errors are clustered at the region level and are robust to heteroskedasticity. This result also holds when excluding ODA targeting emergency response. FE denotes fixed effects; N denotes number.

Sources: OECD based on OECD (2025[1]), CRS: Creditor Reporting System (flows), https://data-explorer.oecd.org/vis?lc=en&pg=0&snb=18&df[ds]; Global Data Lab (2025[2]), “SDG Dashboard”, https://globaldatalab.org/sdgs/.

Is higher DDC ODA associated with improvements in social SDG outcomes?

DDC ODA targeting SDG 1 for poverty reduction is associated with modest but positive improvements in regional income growth. This association also holds when considering country fixed effects. Quantitatively, a one-unit increase in log ODA (per 100 000 people) is associated with a 0.006 increase in the annual change in log GNI per capita (in USD 2017 PPP). This implies that doubling the DDC ODA variable is associated with an additional 0.4 percentage points of annual income growth, pointing to a modest but measurable association (Table A D.2, Columns 1 and 2; see Columns 3 and 4 for regressions based on projects targeting SDGs 1 or 8 on decent work and economic growth).

Table A D.2. Regression results for recipient regions, year-to-year differences model

Notes: Standard errors are clustered at the region level and are robust to heteroskedasticity. FE denotes fixed effects; N denotes number.

Sources: OECD based on OECD (2025[1]), CRS: Creditor Reporting System (flows), https://data-explorer.oecd.org/vis?lc=en&pg=0&snb=18&df[ds]; Global Data Lab (2025[2]), “SDG Dashboard”, https://globaldatalab.org/sdgs/.

The positive and statistically significant association between DDC ODA and income growth is also visible in a simplified long-difference specification that reduces sensitivity to year-to-year volatility. This model regresses the change in log income per capita over the longest period available for each recipient region within 2013-2023 on cumulative DDC ODA per capita over the same period, controlling for baseline income and population. This approach smoothes annual fluctuations in ODA disbursements and income measures and allows more time for outcome variables to adjust (Table A D.3).

Table A D.3. Regression results for recipient regions, long-differences model

Notes: Standard errors are clustered at the region level and are robust to heteroskedasticity. ODA per 1 000 people. FE denotes fixed effects.

Sources: OECD based on OECD (2025[1]), CRS: Creditor Reporting System (flows), https://data-explorer.oecd.org/vis?lc=en&pg=0&snb=18&df[ds]; Global Data Lab (2025[2]), “SDG Dashboard”, https://globaldatalab.org/sdgs/.

Are increases in DDC ODA associated with improvements in environmental SDG outcomes?

Faster growth in DDC ODA for urban sustainability and climate is associated with increases in green areas in cities. In the baseline specification, increases in SDGs 11- and 13-related DDC ODA are linked to faster growth in the share of green space within built-up areas, with a one-unit increase in the annualised change in log ODA per capita associated with a 0.04 percentage point greater annual increase in urban greenness. This relationship is statistically significant at the 5% level. The relationship weakens and loses statistical significance once country fixed effects are included (0.036), although the magnitude remains comparable (Table A D.4, Columns 1 and 2).

The estimates indicate that growth in DDC ODA is linked to reductions in per capita transport CO₂ emissions. In the specification including recipient-country fixed effects, a one-unit increase in the annualised change in log DDC ODA per capita is associated with an annual decrease of 6.8 kilograms of transport CO₂ emissions per person. The coefficient is similar in magnitude to the baseline estimate and statistically significant at the 5% level in both specifications, pointing to a consistent within-country relationship. While transport emissions are shaped by a range of national and local factors, including energy and transport policies, vehicle fleet composition, urban form and infrastructure investment, the results are consistent with the possibility that DDC flows may support emissions reductions at the city level, potentially through urban mobility projects, public transport investment or more sustainable urban infrastructure (Table A D.4, Columns 3 and 4).

Table A D.4. Regression results for recipient cities

Notes: Standard errors are robust to heteroskedasticity. FE denotes fixed effects; N denotes number.

Sources: OECD based on OECD (2025[1]), CRS: Creditor Reporting System (flows), https://data-explorer.oecd.org/vis?lc=en&pg=0&snb=18&df[ds]; Mari Rivero, I. et al. (2025[3]), Urban Centre Database (UCDB) R2024A, https://human-settlement.emergency.copernicus.eu/ucdb2024Overview.php.

A placebo test using DDC ODA commitments rather than disbursements provides additional support that the baseline associations are driven by resources actually received by partner territories. Commitments capture funding that may be implemented and disbursed over several years, and therefore do not necessarily correspond to the timing of observed changes in city-level outcomes. In the placebo specification with recipient-country fixed effects, DDC ODA commitments are not significantly associated with changes in the share of green areas, unlike the baseline specification based on disbursements. For transport-related CO₂ emissions per capita, the placebo coefficients change substantially relative to the baseline results, suggesting that commitments do not reproduce the same empirical pattern observed for disbursed ODA. These results should be interpreted cautiously, but they provide a useful robustness check indicating that the main associations are more closely linked to disbursements than to commitments.

Table A D.5. Regression results for recipient cities, placebo test

Notes: Standard errors are robust to heteroskedasticity. FE denotes fixed effects; N denotes number.

Sources: OECD based on OECD (2025[1]), CRS: Creditor Reporting System (flows), https://data-explorer.oecd.org/vis?lc=en&pg=0&snb=18&df[ds]; Mari Rivero, I. et al. (2025[3]), Urban Centre Database (UCDB) R2024A, https://human-settlement.emergency.copernicus.eu/ucdb2024Overview.php.

Regression results for provider regions

Is higher DDC engagement associated with stronger local governance?

Higher levels of DDC ODA per capita are associated with stronger overall governance performance in OECD Development Assistance Committee (DAC) provider regions. Log ODA per capita is positively and statistically significantly associated with the EQI, controlling for regional gross domestic product (GDP) per capita and population. As the specification includes provider-country fixed effects, this association reflects within-country variation across regions rather than broad differences between countries (Table A D.6, Column 1).

This positive association is also observed for the impartiality, lower corruption and quality of service dimensions of governance. Regions with higher levels of DDC ODA per capita tend to be perceived as having higher impartiality and integrity (lower corruption) in public service provision, and higher perceived quality of services (based on the EQI), with all associations statistically significant (Table A D.6, Columns 2 and 4). While this association may principally result from underlying factors that result in both better governance and more ODA activities, they are also consistent with evidence from the DDC impact surveys, which show that non-financial activities, including peer learning and technical assistance, provide DAC regions and cities with access to new knowledge and innovation while strengthening institutional capacity. DAC regions and cities also report positive perceived impacts of DDC on local governance, particularly in areas related to multi-level governance, inclusiveness and participation, which is consistent with the positive statistical associations between DDC engagement and governance dimensions related with fairness, equal treatment and integrity in public administration. These results are however simple statistical associations and do not provide any evidence about causal effects in one direction or another. In addition to the above-mentioned possibility of this association being spurious, it is also plausible that better-governed regions are more inclined to engage in DDC in the first place.

Table A D.6. Regression results for provider regions

Notes: Standard errors are robust to heteroskedasticity. FE denotes fixed effects; N denotes number.

Sources: OECD based on OECD (2025[1]), CRS: Creditor Reporting System (flows), https://data-explorer.oecd.org/vis?lc=en&pg=0&snb=18&df[ds]; ECOPER (2024[4]), Decentralised Cooperation Report: Mapping Donor Cities, https://www.elankidetza.euskadi.eus/contenidos/documentacion/publicaciones_descentralizada/es_def/INFORME-2024-ENG-.pdf; Charron, N., V. Lapuente and M. Bauhr (2024[5]), “The Geography of Quality of Government in Europe: Subnational Variations in the 2024 European Quality of Government Index and Comparisons with Previous Rounds”, https://www.gu.se/en/quality-government/qog-data/data-downloads/european-quality-of-government-index.

The regression analysis presented in this section is a step forward in quantifying the associations between cross-border DDC ODA and subnational outcome indicators, but the results remain non-causal and should be interpreted with caution. While this represents the first empirical effort to localise provider and recipient CRS data and link it to local governance and SDG outcomes, the analysis is constrained by partial localisation of CRS projects, uneven territorial coverage across countries and levels of government, relatively small samples and unbalanced panels, and limited availability of comparable local governance and SDG outcome indicators. In addition, some estimated relationships are sensitive to the choice of specification and may be influenced by omitted variables and reverse causality.

References

[5] Charron, N., V. Lapuente and M. Bauhr (2024), “The Geography of Quality of Government in Europe: Subnational Variations in the 2024 European Quality of Government Index and Comparisons with Previous Rounds”, QoG Working Paper Series, No. 2024:2, Department of Political Science, University of Gothenburg, https://www.gu.se/en/quality-government/qog-data/data-downloads/european-quality-of-government-index.

[4] ECOPER (2024), Decentralised Cooperation Report: Mapping Donor Cities, https://www.elankidetza.euskadi.eus/contenidos/documentacion/publicaciones_descentralizada/es_def/INFORME-2024-ENG-.pdf.

[2] Global Data Lab (2025), “SDG Dashboard”, https://globaldatalab.org/sdgs/.

[3] Mari Rivero, I. et al. (2025), Urban Centre Database (UCDB) R2024A, European Union, https://human-settlement.emergency.copernicus.eu/ucdb2024Overview.php.

[1] OECD (2025), CRS: Creditor Reporting System (flows), OECD, Paris, https://data-explorer.oecd.org/vis?lc=en&pg=0&snb=18&df[ds].

Notes

← 1.  Country fixed effects ensure that each region is compared to other regions within the same country, removing to some extent the influence of persistent national-level differences such as national institutional frameworks, governance systems or overall development levels. Year fixed effects are also included to account for common trends or shocks affecting all countries simultaneously.

← 2.  In practice, the relevant lag is likely to vary according to the absorptive capacity of recipient territories and the type of intervention. Infrastructure projects, for example, may take longer to affect outcomes than technical assistance. Given the limited number of observations, alternative lag structures could not be tested systematically. This should be revisited as data coverage improves.

← 3.  Conditional convergence means that, after accounting for other relevant characteristics, poorer regions tend to grow faster than richer ones over time, gradually reducing income gaps. Controlling for initial income levels helps ensure that the estimated effects reflect the contribution of cross-border DDC ODA, rather than this natural catch-up process.

← 4.  While the first-difference approach removes the influence of time-invariant characteristics on the level of outcomes, baseline income and population are retained as they may still predict the rate of change of outcomes, through convergence dynamics and scale-related differences in development trajectories respectively.