## Foreign Aid and Conflicts: The Effects of 9/11 on Donor Behavior (Working Paper No. WP/2024/016)

## Source details

**Canonical URL:** [Foreign Aid and Conflicts: The Effects of 9/11 on Donor Behavior (Working Paper No. WP/2024/016)](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025016-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025016-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025016-print-pdf.pdf.json)

---

### Introduction and research question
- Background and motivation
  - Global salience of conflict and terrorism rose sharply after the attacks of September 11, 2001.
  - Authors posit a “scissor effect”: a decline in emphasis on aid effectiveness in the early 2000s coinciding with a rise in donor attention to conflicts and terrorism.
- Primary research question
  - Did the relationship between non-military foreign aid and conflicts (including terror attacks) in recipient countries change after the 9/11 attacks?
- Scope and focus
  - Bilateral, non-military aid commitments (not disbursements or military aid).
  - Time period of analysis: 1980 to 2021.
  - Emphasis on whether donors shifted allocation of non-military aid to support conflict- or terror-affected countries after 9/11.

### Data, sample, and key variables
- Coverage and sample
  - Foreign aid (ODA) data span 1980 to 2021.
  - Focused on the 20 largest donors, which together accounted for 99 percent of aid commitments.
  - Panel: 47-year panel dataset, featuring 189 recipients in total, with an average of 165 recipients per donor-year.
  - Bilateral aid flows exclusively considered; multilateral flows excluded. Bilateral aid constitutes about 75 percent of ODA.
- Main dependent variable
  - ODAdrt = logarithm of commitment ODA from donor d to recipient r at time t.
  - Zero bilateral-aid observations: when ODA = 0 it is assigned the value of one before taking natural logs.
- Main explanatory variables
  - Armed conflicts: UCDP/PRIO Armed Conflict Dataset; onset = year of the first battle-related death; UCDP definition requires at least 25 battle-related deaths.
  - Terror attacks: Global Terrorism Database (GTD); measures include number of killed, number of wounded, number kidnapped, and property damage value (US dollars).
- Political alignment
  - Constructed from UN General Assembly votes (session 49 to session 75) as average fraction of votes concordant with G7 donors.
- Additional controls and alternatives
  - Donor and recipient GDP and population (World Bank WDI).
  - US military aid used as an alternative dependent variable (available as obligations, not commitments).
  - Debt-relief controls: HIPC and MDRI decision/completion dummies.

### Empirical approach and identification
- Baseline specification
  - Gravity-style reduced-form with donor-recipient fixed effects (γdr) to eliminate time-invariant bilateral characteristics.
  - Baseline: ODAdrt = θ CONFLICTrt + X′drt β + γdr + ɛdrt, where CONFLICTrt is either conflict dummy or log of conflict severity.
- Mechanism testing
  - Interaction specification: ODAdrt = θ1 CONFLICTrt + θ2 UNdrt + θ3 CONFLICTrt * UNdrt + X′drt β + γdr + ɛdrt, where UNdrt denotes political alignment.
- Structural break testing and identification
  - Structural break tests detect a single breakpoint in bilateral aid flows; estimated breakpoint date = 2002 at a 99 percent confidence level.
  - Identification leverages timing of conflict onset and terror attacks as plausibly exogenous shocks (strategic surprise).
- Zero-value handling and estimator robustness
  - ODA zeros assigned value 1 then logged.
  - Alternative estimators used in robustness checks: Poisson Pseudo-Maximum Likelihood (PPML), Zero-Inflated Poisson, Heckman selection, Probit.

### Main empirical findings (high level)
- Sign reversal around 2001
  - Full-sample (1980–2021): Armed Conflict (dummy) coefficient = -0.235** (standard error 0.111).
  - Sub-sample pre-2001 (1980–2000): Armed Conflict (dummy) coefficient = -0.723*** (0.134).
  - Sub-sample post-2001 (2001–2021): Armed Conflict (dummy) coefficient = 0.258** (0.127).
  - Quantified impacts (using Columns II and III benchmarks): occurrence of an armed conflict decreases bilateral aid commitment by about 54 percent for 1980–2000; occurrence of an armed conflict increases bilateral aid commitment by about 30 percent for 2001–2021 (source interpretation).
- Drivers of results
  - Results driven by grants rather than loans.
  - Positive post-2001 effects concentrated in purpose-specific aid: humanitarian and social infrastructure.
  - Results robust when substituting armed conflict with terror-attack measures.
- Terrorism measures: sign flip after 2001
  - Number of wounded (log): 1980-2021 = -0.149*** (0.022); 1980-2000 = -0.181*** (0.022); 2001-2021 = 0.041*** (0.015).
  - Number of killed (log): 1980-2021 = -0.175*** (0.022); 1980-2000 = -0.179*** (0.021); 2001-2021 = 0.030* (0.015).
  - Property damage value (log): 1980-2021 = -0.057*** (0.010); 1980-2000 = -0.069*** (0.010); 2001-2021 = 0.017** (0.008).
  - Quantitative elasticity examples post-2001: a 1 percent increase in damages caused by terrorist attacks increases bilateral aid by about 0.04 percent (one specification) and 0.03 percent (another), everything else equal; pre-2001 a 1 percent increase in damages decreased bilateral aid by about 0.18 percent in a reported specification.

### Heterogeneity: conflict types, purpose, and instrument
- Conflict type heterogeneity (Table 3)
  - Internal (intra-state) conflict: 1980-2021 = -0.392*** (0.135); 1980-2000 = -0.721*** (0.168); 2001-2021 = -0.023 (0.153).
  - Internationalized internal conflict: 1980-2021 = 0.171 (0.205); 1980-2000 = -0.504* (0.280); 2001-2021 = 0.862*** (0.224).
  - Interpretation: pre-2001 internal conflicts associated with reductions in ODA; post-2001 internationalized internal conflicts associated with higher ODA.
- Aid composition (post-2003 availability)
  - Purpose aid positive with conflicts: Purpose aid (II) Armed conflict = 0.402*** (0.091).
  - Humanitarian (V): Armed conflict = 0.549*** (0.148).
  - Social infrastructure (IX): Armed conflict = 0.242** (0.102).
  - Quantitatively: occurrence of conflict leads to respectively 28 and 63 percent increases in humanitarian and social infrastructure aid components (source phrasing).
- Grants vs loans (Table 7)
  - Grants: 1980-2021 Armed conflict = -0.239** (0.105); 1980-2000 = -0.722*** (0.129); 2001-2021 = 0.279** (0.121).
  - Loans: coefficients positive but not statistically significant across periods (e.g., 1980-2021 = 0.105 (0.098)).
  - Conclusion: main results driven by grants.

### Political alignment, donor heterogeneity, and debt relief controls
- Political alignment (G7 UN voting concordance)
  - G7 political alignment coefficient overall = -1.085*** (0.237) in selected specifications.
  - Interaction Armed conflict x G7 alignment (2001-2021) = -2.637** (1.298).
  - Interpretation: flip in sign post-2001 is not driven by closer political alignment with G7 countries; higher alignment reduces the positive association between conflict and ODA in post-2001 period.
- Debt-relief initiatives
  - Controlling for HIPC and MDRI decision/completion points does not eliminate the post-2001 sign flip.
- Donor heterogeneity and domestic terrorism experience
  - Donors attacked on their soil after 2001 were more sensitive to the shift in sign.
  - Table 18 examples: Armed Conflict x Years after the first terrorist attack (2001-2021) = 0.937*** (0.161).

### Robustness checks and alternative estimators
- Structural break tests identify 1 break with estimated break year = 2002 (F(1|0) statistic = 57.05; 1% critical value = 12.29).
- Alternative estimators (Table 11)
  - Poisson (1980-2021): Armed conflict (dummy) = -0.022** (0.010).
  - Poisson (1980-2000): -0.077*** (0.015).
  - Poisson (2001-2021): 0.017* (0.010).
  - Probit: pre-2001 negative and significant; post-2001 sign flip present though Probit post-2001 sometimes not statistically significant.
- Regional and exclusion checks
  - Core sign reversal (negative pre-2001; more positive post-2001) persists across many exclusions (excluding US, China, Afghanistan, Iraq, Libya), though post-2001 flips are driven in part by sub-Saharan Africa (SSA) and Middle East and North Africa (MENA) samples.

### Key descriptive and summary statistics (selected)
- Sample sizes and summary statistics (Table 1)
  - ODA: N = 135,276 Mean = 28.3 SD = 425.5; 1980-2000 N = 66,502 Mean = 10.9 SD = 84.1; 2001-2021 N = 68,774 Mean = 45.1 SD = 590.5.
  - Grants: N = 135,276 Mean = 12.1 SD = 83.7; 1980-2000 Mean = 6.9 SD = 60.4; 2001-2021 Mean = 17.2 SD = 101.1.
  - Loans: N = 135,276 Mean = 13.8 SD = 385.0; 1980-2000 Mean = 4.0 SD = 49.9; 2001-2021 Mean = 23.3 SD = 537.5.
  - Armed conflict (recipient/year): N = 7,914 Mean = 0.018 SD = 0.135; 1980-2000 Mean = 0.022 SD = 0.146; 2001-2021 Mean = 0.015 SD = 0.122.
  - Terrorism (recipient/year): N = 7,914 Mean = 0.229 SD = 0.042; 1980-2000 Mean = 0.305 SD = 0.461; 2001-2021 Mean = 0.154 SD = 0.361.
  - Number of wounded (terrorism): N = 7,914 Mean = 23.5 SD = 267.4; 1980-2000 Mean = 20.0 SD = 93.7; 2001-2021 Mean = 27.0 SD = 366.1.
  - Political alignment (G7 voting agreement): N = 4,336 Mean = 0.53 SD = 0.13; 1980-2000 Mean = 0.57 SD = 0.11; 2001-2021 Mean = 0.52 SD = 0.14.
  - Recipient GDP (millions USD): N = 6,587 Mean = 958,534 SD = 559,975; 1980-2000 Mean = 32,429 SD = 90,430; 2001-2021 Mean = 148,382 SD = 748,652.
  - Donor GDP (millions USD): N = 816 Mean = 1,600,281 SD = 2,961,620; 1980-2000 Mean = 884,907 SD = 1,567,202; 2001-2021 Mean = 2,288,141 SD = 3,727,784.

### Policy implications and research agenda
- Policy implications highlighted by authors
  - A fundamental rethink of the aid industry is suggested: recipient needs should take precedence over donor preference.
  - Mechanisms to elicit recipient-country needs in ways that align with taxpayers in donor countries should be a priority.
  - Aid effectiveness debates must remain central amid new priorities; allocative and technical efficiency of foreign aid should be complementary.
  - In a world increasingly affected by conflicts (including related to climate change), aid should both increase and be more effective to support countries facing existential threats.
- Research agenda
  - Further research should explore the effectiveness of foreign aid in conflict-afflicted countries, addressing acute endogeneity by building appropriate counterfactuals to test whether situations would have been worse without donors’ interventions.

*Source: IMF Working Papers — Foreign Aid and Conflicts: The Effects of 9/11 on Donor Behavior (Working Paper No. WP/2024/016).*

### 1. Introduction ........................................................................................................

### 1. Introduction

### Background and motivation
- The global salience of conflict and terrorism rose sharply after the attacks of September 11, 2001.
- Accounts from aid specialists and data on phrase usage (e.g., “fragile states”) suggest 9/11 was a turning point in donors’ willingness to support nations grappling with conflict and terror attacks.
- Prior to 2001, development aid emphasis was primarily on governance and corruption, reflecting a focus on maximizing aid effectiveness.
- The authors posit a “scissor effect”: a decline in emphasis on aid effectiveness in the early 2000s coinciding with a rise in donor attention to conflicts and terrorism.

### Research question and scope
- Primary question: Did the relationship between non-military foreign aid and conflicts (including terror attacks) in recipient countries change after the 9/11 attacks?
- The paper focuses on bilateral, non-military aid commitments (not disbursements or military aid).
- Time period of analysis: 1980 to 2021.
- The authors are specifically not interested in military aid; they test whether donors shifted allocation of non-military aid to support conflict- or terror-affected countries after 9/11.

### Data sources and key dataset features
- Armed conflicts: UCDP/PRIO Armed Conflict Dataset (International Peace Research Institute, Oslo).
- Terror attacks: Global Terrorism Database (University of Maryland).
- Aid data: Official development assistance (ODA) data provided by members of the Development Assistance Committee (DAC) of the OECD.
- Aid focus: exclusively bilateral aid (as opposed to multilateral aid), which accounts for about 75 percent of overall aid globally.
- Rationale for using commitments rather than disbursements: aid is disbursed several years after commitment; using commitments helps address endogeneity concerns between aid flows and conflict onset.

### Empirical approach
- Main estimation: gravity model of bilateral donor-recipient aid flows saturated with donor-recipient dummies, akin to a difference-in-difference model.
- Identification strategy: exploit the timing of onset of domestic armed conflicts and domestic terror attacks as plausibly exogenous shocks (strategic element of surprise).
- Structural break test: employ a simple sample split at year 2001 to test for a change in the relationship after 9/11.
- Additional design choices: focus on aid commitments, exclude multilateral aid to maintain donor attribution clarity.

### Main findings (high-level)
- The sign of the causal relationship linking bilateral aid commitments to the onset of armed conflicts in recipient countries is statistically significant and goes from negative to positive after the year 2001.
- Results are driven by:
  - Grants rather than loans.
  - Aid for health and humanitarian purposes.
- Findings are robust to:
  - Substituting armed conflict with terror attacks.
  - Accounting for debt relief initiatives.
  - Different estimators.
- Interpretation: results are consistent with a shift in donors’ preferences toward supporting fragile states following 9/11, suggesting donors’ preferences took primacy over recipient needs.

### Relation to existing literature
- Connects two strands: aid allocation/effectiveness literature and literature on conflicts/terrorism.
- Notes historical emphasis:
  - Literature on aid effectiveness peaked in early 2000s (e.g., Burnside and Dollar, 2000).
  - Literature on conflicts and terrorism gained prominence thereafter.
- Links to recent findings: evidence that donors shifted toward badly governed countries, worsening the “paradox of aid.”
- Methodological kinship: follows natural experiment approaches used in aid literature (e.g., timing exploits similar to Faye and Niehaus (2012)).

### Structure of the paper
- Section 2: Data.
- Section 3: Empirical strategy.
- Section 4: Results, including:
  - 4.1 Did the relationship between foreign aid and conflicts change after 9/11 attacks?
  - 4.2 Did the relationship between foreign aid and terror attacks also change after 9/11?
  - 4.3 How did the composition of aid evolve following conflicts?
  - 4.4 Is political alignment driving the relationship between aid and conflict?
- Section 5: Robustness checks.
- Section 6: Conclusion.

*IMF WORKING PAPERS — Foreign Aid and Conflicts: The Effects of 9/11 on Donor Behavior*

### 2. Data

### 2. Data

### Data coverage and sample
- Foreign aid (ODA) data span the period from 1980 to 2021.
- Focused on the 20 largest donors during the study period, which together accounted for 99 percent of aid commitments.
- Resulting panel: a 47-year panel dataset, featuring 189 recipients in total, with an average of 165 recipients per donor-year.
- Bilateral aid flows are exclusively considered; multilateral flows are excluded.
- ODA includes grants or loans for economic development and welfare enhancement, incorporates technical cooperation, and excludes grants, loans, and credits for military purposes (the ODA data exclude military aid).
- China’s role as a non-traditional creditor donor is noted via AidData; removing China does not alter main results (statement in source).

### Main dependent variable: ODA
- ODA is the main dependent variable; in estimations ODAdrt is the logarithm of commitment ODA from donor d to recipient r at time t.
- Bilateral aid constitutes 75 percent of ODA (interpretation note: results should be seen as an average—on impact—response of the main bilateral donors).
- Zero bilateral-aid observations are not dropped. When ODA = 0 it is assigned the value of one before taking natural logs.

### Main explanatory variables
- Armed conflict (UCDP/PRIO Armed Conflict Dataset):
  - Timing refers to onset = year of the first battle-related death.
  - UCDP definition: “a contested incompatibility that concerns government and/or territory where the use of armed force between two parties, of which at least one is the government of a state, results in at least 25 battle-related deaths.”
  - UCDP captures occurrence and severity of armed conflicts of all types.
- Terror attacks (Global Terrorism Database, GTD, University of Maryland):
  - GTD contains number of killings, number of persons wounded, hostages, and estimates of property damage value.
  - GTD definition: threatened or actual use of illegal force and violence by a nonstate actor to attain a political, economic, religious, or social goal through fear, coercion, or intimidation.
  - GTD is an open-source database with information on over 200,000 terrorist events; information added only when sources are deemed credible.

### Political alignment measure
- Constructed from UN General Assembly votes dataset (session 49 (1995/1996) to session 75 (2020/2021)) compiled by Voelsen et al. (2021).
- Alignment measure: average fraction of UN GA votes for which the recipient country and the Group of 7 (G7) donor countries cast concordant votes (yes, no, or abstention).
- Calculated for each year between a specific G7 donor and recipient, then averaged across all G7 donors to determine the recipient’s alignment.
- Votes from absent or non-member countries are excluded, as are proposals with no voting mechanism.

### Alternative dependent and additional controls
- US military aid is used as an alternative dependent variable (data limitations: available data are on US military aid obligations rather than commitments).
- Demographic and economic control variables obtained from the World Bank’s World Development Indicators database (World Bank, 2023), including population and Gross Domestic Product (GDP) for donor and recipient countries.
- Step dummies and Multilateral Debt Relief Initiative (MDRI) indicators are included to account for decision/completion points related to debt relief initiatives (Heavily Indebted Poor Countries Initiative (HIPIC) referenced).

### Methodological and estimator choices (as applied to the data)
- Baseline reduced-form specification follows Faye and Niehaus (2012); donor-recipient fixed effects (γdr) are included to eliminate time-invariant bilateral characteristics.
- Formal model (Equation (1) as presented in source):
  - ODAdrt = θ CONFLICTrt + X′drt β + γdr + ɛdrt
  - Where CONFLICTrt is either a dummy (1 if country r has any conflicts in year t) or the logarithm of the total number of persons killed, injured, taken hostage, or estimate of property value in US dollars at time t.
  - X′drt is a vector of time-varying donor or recipient control variables (e.g., GDP, population).
- Mechanism testing via interaction (Equation (2) as presented in source):
  - ODAdrt = θ1 CONFLICTrt + θ2 UNdrt + θ3 CONFLICTrt * UNdrt + X′drt β + γdr + ɛdrt
  - UNdrt denotes the political alignment measure.
- Zero-value handling and estimator robustness:
  - Because many donor–recipient pairs have zero aid, the paper assigns ODA = 1 for zero observations and logs the series.
  - Alternative robust estimators employed for robustness checks include Poisson Pseudo-Maximum Likelihood (PPML), Zero-Inflated Poisson, Heckman selection model, and Probit model (references: Silva and Tenreyro, 2006; Herrera, 2010; Martin and Hall, 2017).
- Identification robustness:
  - Donor-recipient fixed effects remove time-invariant confounders.
  - Time-varying controls (population, GDP) mitigate concurrent trends in aid and conflicts.
  - Clustering of standard errors is performed on donor-recipient pairs (Bertrand et al., 2004) to address panel inference concerns.
- Structural break testing:
  - Tests for multiple structural breaks in panel data (Bai and Perron (1998), Ditzen et al. (2021), Karavias et al. (2021)) confirm the existence of a single breakpoint in bilateral aid flows.
  - The estimated breakpoint date is 2002, at a 99 percent confidence level (see Appendix Table A1 in source).

### Key empirical findings (data-driven results summarized in source)
- Full-sample (1980–2021) regression of Equation (1) with a conflict dummy shows a negative and statistically significant relationship between aid commitment and the occurrence of armed conflict (Column I, Table 2 in source).
- Sub-sample analysis:
  - 1980–2000: statistically significant and negative relationship between aid commitment and occurrence of conflicts (Columns II in Table 2).
  - 2001–2021: statistically significant and positive relationship between aid commitment and occurrence of conflicts (Column III in Table 2).
- Quantification (using Columns II and III as benchmarks):
  - Occurrence of an armed conflict decreases bilateral aid commitment by about 54 percent for the sub-sample period before 2001.
  - Occurrence of an armed conflict increases bilateral aid commitment by about 30 percent for the sub-sample period after 2001.
  - Source caveat: these quantifications pertain to change in aid commitments on impact (interpretation note in source).
- Conflict intensity:
  - When replacing the conflict dummy with major/minor conflict dummies (Columns IV–VI, Table 2), the sign flip before vs. after 2001 remains.
  - Results are driven by major conflicts; coefficients suggest the effect of major conflicts are multiple the size in absolute value than that for overall armed conflicts.
  - The effect of major conflicts on aid commitment could increase by over 1 percent after 2001 (source phrasing).
- Robustness statements in source:
  - Main results hold when controlling for donor-year fixed effects or year fixed effects.
  - Results are robust to exclusion of the US and/or China as donors and to exclusion of Afghanistan, Iraq, and Libya as recipients.
  - Donor countries that were attacked on their soil after 2001 were more sensitive to the shift in sign.

*Source: IMF Working Papers — Foreign Aid and Conflicts: The Effects of 9/11 on Donor Behavior (content unit: 2. Data).*

### 4.2 Did the relationship between foreign aid and terror attacks also change after

### wpiea2025016-print-pdf - 4.2 Did the relationship between foreign aid and terror attacks also change after

### Effect of conflict types on aid commitments
- PRIO-defined armed conflicts are contested incompatibilities resulting in at least 25 battle-related deaths.
- PRIO conflict types considered: intra-state, inter-state, and externalized (conflicts involving at least a third party).
- Empirical results (Table 3):
  - Coefficients for intra-state conflicts: statistically significant and negative for the overall sample and pre-2001; negative but not significant post-2001.
  - Coefficients for inter-state conflicts: not statistically significant across overall, pre-2001, and post-2001.
  - Coefficients for internationalized (externalized) conflicts: positive but not statistically significant for the overall sample.
  - Internalized conflicts: statistically significant and negative for pre-2001, statistically significant and positive for post-2001.
- Interpretation: main results are driven by intra-state and internationalized conflicts, suggesting terrorist attacks and international involvement in conflicts could be important drivers of foreign aid allocation.

### Substituting terrorist-attack measures for conflict dummies — sign flip after 2001
- Terrorist-attack measures used: number of persons wounded and killed; persons kidnapped; value of property damage (US dollars).
- Results (Tables 4 and 5):
  - Coefficients associated with terrorist attacks remain statistically significant across specifications.
  - The sign of coefficients associated with terrorist attacks flips after 2001.
  - Quantitative interpretation for post-2001 (columns III and VI): a 1 percent increase in damages caused by terrorist attacks increases bilateral aid by about 0.04 percent and 0.03 percent respectively, everything else equal.
  - For the period pre-2001: a 1 percent increase in damages caused by terrorist attacks decreases bilateral aid by about 0.18 percent.
  - Results in Table 5 are consistent with main results; estimates are interpreted as an average response of main bilateral donors to terrorist attacks on impact.
- Time-evolution check:
  - Regression coefficients plotted using 10-year rolling windows for all four measures (Figure 5) confirm coefficients are negative before progressively turning positive as the rolling window moves past 2001.
  - The flip in sign after the year 2001 is a robust feature of the relationship between terrorist attacks and aid commitments.

### Composition of aid following conflicts (post-2003 data)
- Data on composition of aid commitment available starting in 2003; regression results for post-2003 period reported in Table 6.
- Key findings (Columns I-IX, Table 6):
  - Non-purpose aid commitments: statistically significant and negative relationship with conflicts.
  - Purpose aid commitments: statistically significant and positive relationship with conflicts.
  - Disaggregated purpose-aid drivers: humanitarian and social infrastructure components drive the positive and statistically significant results.
  - Quantitatively: the occurrence of conflict leads to respectively 28 and 63 percent increases in humanitarian and social infrastructure aid components.
  - Debt reduction component decreases following conflicts, suggesting debt initiatives are not driving the increase in purpose aid.

### Grants versus loans
- Table 7 contrasts grants (Columns I-III) and loans (Columns IV-VI) over overall, pre-2001, and post-2001 periods.
  - Grants: coefficients associated with conflicts are statistically significant across the board and flip sign after 2001.
  - Loans: coefficients associated with conflicts are positive across the board but not statistically significant.
- Conclusion: main results are driven by grants rather than loans. Donors increased use of grants in conflict contexts after 9/11; conflicts-afflicted countries are less able to acquire loans. Debt relief initiatives do not appear to confound the post-2001 sign flip.

### Political alignment (UN voting patterns) and aid response
- Regression using Equation (2) includes an interaction between conflicts and political alignment with G7 countries (Columns I-III, Table 8).
  - Coefficient associated with conflicts flips sign; results are statistically significant only for the period after 2001.
  - Interaction terms between conflicts and political alignment with G7 countries are statistically significant but negative in the post-2001 period.
  - Interpretation: flip in sign post-2001 is not driven by political alignment with G7 countries; G7 countries do not increase aid in conflict-afflicted countries that have closer political alignment with them.
  - Note: political alignment measure (closedness in UN voting patterns) is an imperfect proxy of strategic alignment, especially when political turnover accompanies conflicts.
  - Authors note qualitatively similar results when restricting political alignment to US only (not shown).

### Robustness checks and extensions
- Debt-relief initiatives:
  - Accounted for HIPC and MDRI decision and completion points via country-specific step dummies (decision/completion dates).
  - Only 36 countries concerned by these initiatives, of which 33 are African countries.
  - Table 9: main results (flip in sign after 2001) are robust to controlling for debt initiatives.
- US military versus development aid (Table 10):
  - Development aid: confirms flip in sign, though coefficient for conflicts not statistically significant post-2001.
  - US military aid: negative relationship with conflicts but not statistically significant; military aid data are obligations and not directly comparable to commitments.
  - Interpretation: military aid does not follow the development-aid pattern of increasing at onset of conflicts after 2001; data comparability and identification issues noted.
- Alternative estimators for many zeros (Table 11):
  - Estimators used: OLS (benchmark), Poisson Pseudo-Maximum Likelihood, Probit.
  - Coefficients indicate flip in sign after 2001 for Poisson and Probit estimators, though Probit coefficient not statistically significant.
  - Main results robust to estimator choice.
- Regional exclusion checks (Tables 12–14):
  - Table 12 (overall period): aid commitments are negatively affected by onset of conflicts across regressions excluding different regions, except when excluding Europe and Central Asia.
  - Table 13 (pre-2001): relationship between aid commitment and conflicts is negative across the board.
  - Table 14 (post-2001): flip in sign of conflict coefficient is not robust to excluding sub-Saharan Africa and the Middle East and North Africa separately.
  - Interpretation: sub-Saharan Africa and Middle East and North Africa regions drive results; both regions have experienced many conflicts.

### Conclusion and policy implications
- Main empirical conclusions:
  - The sign of the relationship linking bilateral aid commitments to the onset of armed conflicts in recipient countries goes from negative to positive after the year 2001.
  - Results are driven by grants rather than loans and by aid for health and humanitarian purposes.
  - Findings are robust to substituting armed conflicts with terror attacks, accounting for debt relief initiatives, and alternative estimators.
- Interpretation: results stem from a shift in donors’ preferences toward supporting fragile states following the 9/11 attacks, confirming the primacy of donors’ preference over recipient needs.
- Policy implications and recommendations:
  - A fundamental rethink of the aid industry is suggested: recipient needs should take precedence over donor preference.
  - Mechanisms to elicit recipient-country needs in ways that align with taxpayers in donor countries should be a priority.
  - Aid effectiveness debates must remain central amid new priorities; allocative and technical efficiency of foreign aid should be complementary.
  - In a world increasingly affected by conflicts (including related to climate change), aid should both increase and be more effective to support countries facing existential threats.
- Research agenda:
  - Further research should explore the effectiveness of foreign aid in conflict-afflicted countries, addressing acute endogeneity by building appropriate counterfactuals to test whether situations would have been worse without donors’ interventions.

*Italic: Source — IMF Working Paper section "4.2 Did the relationship between foreign aid and terror attacks also change after", from the document provided.*

### 2009. doi: 10.1080/01402390802407392.

### wpiea2025016-print-pdf - 2009. doi: 10.1080/01402390802407392.

### Citation
- Authors: Todd Moss, David Roodman, and Scott Standley.
- Title: The global war on terror and u.s. development assistance: USAID allocation by country, 1998-2005.
- Outlet: Center for Global Development, Working Papers, 01.
- Year and identifier: 2009. doi: 10.1080/01402390802407392.

### Document scope (as provided)
- Focus: USAID allocation by country, 1998-2005 (as stated in the title).
- Publication type: Working Paper (Center for Global Development, Working Papers, 01).

*Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025016-print-pdf.pdf*

### 2005. doi: 10.2139/ssrn.984258.

### Foreign Aid and Conflicts: The Effects of 9/11 on Donor Behavior (Working Paper No. WP/2024/016)

### Trends and descriptive statistics
- Google Books trends: "fragile state" rebased at 100 for the year 2001 shows a rise; "development aid effectiveness" shows a fall (three-year smoothing, American English; robust across English, French, Spanish).
- Evolution of bilateral aid commitments shown for 1980 to 2021: total bilateral aid commitments and separate series for recipients with highest (strictly above the third quintile) and lowest (strictly below the first quintile) conflict frequency.
- Country-level examples: time series displayed for Nigeria, Libya, Afghanistan, and Iraq — bilateral aid commitments (blue), terrorist attacks (orange), and armed conflicts (green). Data sources: OECD DAC ODA, UCDP/PRIO Armed Conflict Dataset, Global Terrorism Database (University of Maryland).
- Table 1 (Descriptive Statistics, 1980-2021; 1980-2000; 2001-2021):
  - ODA: N=135,276 Mean=28.3 SD=425.5; 1980-2000 N=66,502 Mean=10.9 SD=84.1; 2001-2021 N=68,774 Mean=45.1 SD=590.5.
  - Grants: N=135,276 Mean=12.1 SD=83.7; 1980-2000 Mean=6.9 SD=60.4; 2001-2021 Mean=17.2 SD=101.1.
  - Loans: N=135,276 Mean=13.8 SD=385.0; 1980-2000 Mean=4.0 SD=49.9; 2001-2021 Mean=23.3 SD=537.5.
  - Armed conflict (recipient/year): N=7,914 Mean=0.018 SD=0.135; 1980-2000 Mean=0.022 SD=0.146; 2001-2021 Mean=0.015 SD=0.122.
  - Terrorism (recipient/year): N=7,914 Mean=0.229 SD=0.042; 1980-2000 Mean=0.305 SD=0.461; 2001-2021 Mean=0.154 SD=0.361.
  - Number of wounded (terrorism): N=7,914 Mean=23.5 SD=267.4; 1980-2000 Mean=20.0 SD=93.7; 2001-2021 Mean=27.0 SD=366.1.
  - Number of killed (terrorism): N=7,914 Mean=25.8 SD=273.1; 1980-2000 Mean=27.8 SD=163.6; 2001-2021 Mean=23.8 SD=349.6.
  - Property damage value (terrorism): N=6,618 Mean=0.40 SD=10.2; 1980-2000 Mean=0.50 SD=12.3; 2001-2021 Mean=0.31 SD=7.8.
  - Political alignment (G7 voting agreement): N=4,336 Mean=0.53 SD=0.13; 1980-2000 Mean=0.57 SD=0.11; 2001-2021 Mean=0.52 SD=0.14.
  - Recipient GDP (millions USD): N=6,587 Mean=958,534 SD=559,975; 1980-2000 Mean=32,429 SD=90,430; 2001-2021 Mean=148,382 SD=748,652.
  - Donor GDP (millions USD): N=816 Mean=1,600,281 SD=2,961,620; 1980-2000 Mean=884,907 SD=1,567,202; 2001-2021 Mean=2,288,141 SD=3,727,784.

### Main empirical findings: foreign aid and armed conflict
- Table 2: Bilateral aid regressions (dependent variable: log bilateral ODA); donor-recipient fixed effects (DR); standard errors clustered at donor-recipient pair level.
  - 1980-2021 (Column I): Armed Conflict (dummy) coefficient = -0.235** (standard error 0.111).
  - 1980-2000 (Column II): Armed Conflict (dummy) coefficient = -0.723*** (0.134).
  - 2001-2021 (Column III): Armed Conflict (dummy) coefficient = 0.258** (0.127).
  - Major Armed Conflict (2001-2021, Column V/VI variants): Major armed conflict (dummy) 1.061*** (0.281) in specific specifications.
- Interpretation: Negative association between armed conflict and donor bilateral ODA in 1980-2000; relationship reverses to positive in the 2001-2021 period (post-9/11 era), with statistical significance in key specifications.

### Conflict type heterogeneity
- Table 3: Types of armed conflict (internal, interstate, internationalized internal) — bilateral aid regressions.
  - 1980-2021 (Column I): Internal armed conflict (dummy) = -0.392*** (0.135); Interstate armed conflict (dummy) = -0.175 (0.209); Internationalized internal conflict (dummy) = 0.171 (0.205).
  - 1980-2000 (Column II): Internal = -0.721*** (0.168); Internationalized internal = -0.504* (0.280).
  - 2001-2021 (Column III): Internal = -0.023 (0.153); Internationalized internal = 0.862*** (0.224).
- Interpretation: Pre-2001 negative effects of internal conflict on ODA; post-2001 internationalized internal conflicts associated with higher ODA.

### Terrorism measures and aid
- Table 4: Terrorism fatalities and injuries (log) on bilateral aid.
  - Number of wounded (log): 1980-2021 = -0.149*** (0.022); 1980-2000 = -0.181*** (0.022); 2001-2021 = 0.041*** (0.015).
  - Number of killed (log): 1980-2021 = -0.175*** (0.022) in alternate specification; 1980-2000 = -0.179*** (0.021); 2001-2021 = 0.030* (0.015).
- Table 5: Kidnappings and property damages (log) on bilateral aid.
  - Number of kidnapped (log): 1980-2021 = -0.064** (0.025); 1980-2000 = -0.153*** (0.026); 2001-2021 = 0.097*** (0.020).
  - Property damage value (log): 1980-2021 = -0.057*** (0.010); 1980-2000 = -0.069*** (0.010); 2001-2021 = 0.017** (0.008).
- Interpretation: Terrorism-related severity indicators negatively associated with bilateral ODA pre-2001, but coefficients flip positive and significant in 2001-2021 in multiple measures.

### Aid composition: purposes, grants vs loans
- Table 6: Purpose vs non-purpose aid categories (52,791 observations; fixed effects DR).
  - Armed conflict associations (selected columns):
    - Non-purpose aid (I): Armed conflict = -0.152 (0.107).
    - Purpose aid (II): Armed conflict = 0.402*** (0.091).
    - Humanitarian (V): Armed conflict = 0.549*** (0.148).
    - Social infrastructure (IX): Armed conflict = 0.242** (0.102).
  - Interpretation: Recipient armed conflict associated with increases in purpose-specific aid categories (notably humanitarian and social infrastructure) while non-purpose aid not significantly positive.
- Table 7: Grants vs loans
  - Bilateral grants (I, 1980-2021): Armed conflict = -0.239** (0.105).
  - Bilateral grants (II, 1980-2000): Armed conflict = -0.722*** (0.129).
  - Bilateral grants (III, 2001-2021): Armed conflict = 0.279** (0.121).
  - Bilateral loans (IV-VI): Armed conflict coefficients generally small and not significant (e.g., 0.105 (0.098) for 1980-2021).
  - Interpretation: Shifts toward using grants in post-2001 period where grants are positively associated with recipient armed conflict; loans show weaker relationships.

### Political alignment, debt relief, and donor heterogeneity
- Table 8: Accounting for G7 political alignment (1995-2021 sample):
  - G7 political alignment coefficient = -1.085*** (0.237) overall; splits show 2.155*** (0.677) for 1995-2000 and -0.980*** (0.231) for 2001-2021.
  - Interaction Armed conflict x G7 political alignment (2001-2021) = -2.637** (1.298).
  - Interpretation: Political alignment with G7 donors matters strongly; in 2001-2021, higher alignment reduces the positive association between recipient armed conflict and ODA.
- Table 9: Controlling for HIPC and MRDI (debt relief decision/completion points) — Armed conflict coefficients remain similar (e.g., 1980-2021 = -0.234** (0.109); 2001-2021 = 0.259** (0.127)).

### US aid distinctions and military vs development aid
- Table 10: US bilateral aid and US military aid (1980-2019, 1980-2000, 2001-2019 samples; observations = 6,079 total):
  - US bilateral aid (I-III): Armed conflict (dummy) = -0.507 (0.540) for 1980-2019; -1.367** (0.581) for 1980-2000; 0.445 (0.433) for 2001-2019.
  - US military aid (IV-VI): Armed conflict coefficients not consistently significant (e.g., -0.400 (0.578) for 1980-2019; -0.101 (0.612) for 1980-2000; -0.378 (0.748) for 2001-2019).
  - Interpretation: US bilateral aid shows pre-2001 reductions associated with armed conflict; post-2001 positive coefficient in some specifications for total US bilateral aid, while US military aid displays no consistent positive shift.

### Robustness checks and alternative estimators
- Structural break tests (Table A1): Sequential test for multiple breaks in bilateral aid data identifies 1 break with estimated break year = 2002 (F(1|0) statistic = 57.05; 1% critical value = 12.29).
- Alternative estimators for "too many zeros" (Table 11):
  - Poisson (1980-2021): Armed conflict (dummy) = -0.022** (0.010).
  - Poisson (1980-2000): Armed conflict = -0.077*** (0.015).
  - Poisson (2001-2021): Armed conflict = 0.017* (0.010).
  - Probit (1980-2021): Armed conflict = -0.170*** (0.042).
  - Probit (1980-2000): -0.296*** (0.059).
  - Probit (2001-2021): 0.093 (0.072).
  - Interpretation: Results are robust to Poisson and Probit specifications; sign reversal across periods remains.

### Regional and exclusion robustness
- Excluding regions (Tables 12–14) yields:
  - 1980-2021 excluding MENA: Armed conflict = -0.497*** (0.121); excluding SSA: -0.606*** (0.165); excluding NA: -0.234** (0.111); excluding ECA: 0.225** (0.105).
  - 1980-2000 exclusions produce consistently negative Armed conflict coefficients (e.g., overall -0.701*** to -1.069*** across regions).
  - 2001-2021 exclusions produce mostly positive Armed conflict coefficients (e.g., overall 0.249* (0.133); excluding ECA 0.467*** (0.134)).
- Excluding key countries (Tables 16–20):
  - Excluding China (Table 16): 1980-2021 Armed conflict = -0.252** (0.112); 2001-2021 = 0.242* (0.127).
  - Excluding Afghanistan, Iraq, Libya (Table 17): 1980-2021 Armed conflict = -0.342*** (0.112); 2001-2021 = 0.140 (0.127).
  - Excluding United States (Table 19): 1980-2021 Armed conflict = -0.220* (0.113); 2001-2021 = 0.252* (0.133).
  - Excluding US, China, Afghanistan, Iraq, Libya (Table 20): 1980-2021 Armed conflict = -0.337*** (0.115); 2001-2021 = 0.124 (0.132).
- Interpretation: Core results — negative association pre-2001 and more positive association post-2001 — persist across multiple exclusions and regional samples, though magnitudes and significance vary.

### Temporal dynamics and donor experience of terrorism
- Table 18: Differentiated effects before and after the first terrorist attack in donor countries:
  - For 1980-2000 period:
    - Armed Conflict x Years before the first terrorist attack (1980-2000) = -2.033** (0.905).
    - Armed Conflict x Years after the first terrorist attack (1980-2000) = -1.116*** (0.163).
  - For 2001-2021 period:
    - Armed Conflict x Years before the first terrorist attack (2001-2021) = 0.399 (0.533).
    - Armed Conflict x Years after the first terrorist attack (2001-2021) = 0.937*** (0.161).
  - Interpretation: Donor domestic experience with terrorism changes the donor response to recipient armed conflict; post-2001 years after donor terrorist attacks are associated with increased aid to conflicted recipients.

### Time fixed effects and overall robustness
- Table 15: Including year fixed effects (DR-Y):
  - 1980-2021 Armed conflict (dummy) = -0.159 (0.108).
  - 1980-2000 Armed conflict = -0.552*** (0.128).
  - 2001-2021 Armed conflict = 0.211* (0.127).
  - Major armed conflict (2001-2021) = 1.187*** (0.278) in specific specifications.
- Structural break detection and rolling-window coefficient plots:
  - Sequential break tests detect a structural break estimated in 2002.
  - Rolling 10-year windows for terrorism and armed conflict coefficients show evolving relationships; figures display point estimates and 95% confidence intervals.

### Key analytical takeaways
- Pre-2001 (1980-2000): Recipient armed conflict and higher terrorism metrics are generally associated with lower bilateral ODA and grants; internal conflicts lead to reductions in aid.
- Post-2001 (2001-2021): The relationship changes — recipient armed conflict and terrorism severity measures are often associated with higher bilateral ODA, especially for purpose-specific aid (notably humanitarian), grants, and in cases of internationalized internal conflict.
- Donor characteristics and politics matter: donor GDP, donor population, and G7 political alignment are significant covariates; political alignment reduces the positive post-2001 association between conflict and aid.
- Robustness: Results hold across multiple estimators (Poisson, Probit), regional exclusions, country exclusions (US, China, Afghanistan, Iraq, Libya), and after controlling for debt relief decision/completion points. Structural break tests identify a break around 2002.

*Source: IMF Working Papers — Foreign Aid and Conflicts: The Effects of 9/11 on Donor Behavior, Working Paper No. WP/2024/016.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025016-print-pdf.pdf_