## wpiea2022046-print-pdf - References

## Source details

**Canonical URL:** [wpiea2022046-print-pdf - References](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022046-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022046-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022046-print-pdf.pdf.json)

---

### I. Introduction: context and motivation
- DSSI and COVID-19 response:
  - Since May 2020, the G20’s Debt Service Suspension Initiative (DSSI) provided eligible low-income countries a temporary moratorium on official bilateral debt repayments.
- China’s role and scale:
  - More than US$350 billion in official finance committed to developing countries between 2000 and 2014 (Dreher et al., 2017).
  - In Africa, Beijing is believed to have signed some 1,141 loan commitments worth US$153 billion over 2000-19.
  - China launched the US$1 trillion Belt and Road Initiative (BRI) in 2013.
  - China’s share of bilateral debt owed by the world’s poorest countries to the G20 rose from 45 per cent in 2015 to 63 percent in 2019.
  - As of October 2020, China postponed over US$1.9 billion in repayments under the DSSI, out of roughly US$5.3 billion suspended by G20 members for 44 debtor countries.
- Data transparency constraints:
  - China does not publish detailed country- and project-level foreign aid data; aid is considered “a sensitive area [and] a state secret”.
  - China State Council’s White Paper (January 2021) reported RMB 270.2 billion (US$ 41.6 billion) in aid between 2013-18 without granular breakdowns.
- AidData contribution:
  - AidData’s Global Chinese Official Finance Dataset captures more than 4,000 Chinese government-financed projects in 138 countries over 2000-14 using the TUFF methodology.

### II. Research objective and approach
- Purpose:
  - Take stock of conflicting empirical literature on Chinese aid effectiveness.
  - Employ meta-regression analysis (MRA) using 473 estimates from 15 studies that account for endogeneity.
- MRA aims:
  - (i) examine presence of publication selection bias in the literature;
  - (ii) quantify the genuine effect of Beijing’s official finance on developing countries;
  - (iii) explain heterogeneity across reported results.
- Literature screening:
  - Initial search identified 190 studies; search ended in December 2020.
  - Exclusions: non-English papers; unpublished “grey” literature; unrelated papers; descriptive/qualitative studies without quantitative estimates.

### III. Meta-dataset construction and coding
- Sample composition:
  - Final meta-dataset: 15 studies and 473 estimates (Appendix Table A1).
  - Inclusion criteria: studies addressing identification and providing regression-based estimates; excluded estimates not accounting for endogeneity.
- Collected/coded for each study:
  - sample size, regression coefficients, standard errors or t-statistics; paper metadata; dependent-variable category; measurement of Chinese aid; sample and model specification.
- Dependent-variable categories:
  - (i) “economic development”; (ii) “social development”; (iii) “governance”; (iv) “stability”; (v) “other foreign finance”; (vi) “soft power”.
- Key descriptive statistics (N = 473 observations):
  - Partial correlation: Mean 0.001; S.D. 0.076.
  - SE: Mean 0.021; S.D. 0.016.
  - Author affiliation: Non-Chinese mean 0.753; S.D. 0.432. Chinese mean 0.247; S.D. 0.432.
  - Publication outlet: Working paper mean 0.581; S.D. 0.494. Peer-reviewed journal mean 0.419; S.D. 0.494.
  - Measurement of Chinese aid: All or non-ODA flows mean 0.552; S.D. 0.498. ODA flows only mean 0.448; S.D. 0.498.
  - Aggregate vs sector-specific: Sector-specific mean 0.201; S.D. 0.423. Aggregate mean 0.799; S.D. 0.423.
  - Variable form: Continuous mean 0.630; S.D. 0.483. Dummy mean 0.218; S.D. 0.413. Number of projects mean 0.152; S.D. 0.360.
  - Model specification: Without other donors mean 0.937; S.D. 0.244. Other donors mean 0.063; S.D. 0.244.
  - Sample level: Micro mean 0.638; S.D. 0.481. Macro mean 0.362; S.D. 0.481.
  - Region: Worldwide mean 0.307; S.D. 0.462. Africa mean 0.651; S.D. 0.477. LAC/Asia mean 0.042; S.D. 0.201.
  - Recipients' outcome incidence (means): Economic development mean 0.256; S.D. 0.437. Governance mean 0.068; S.D. 0.251. Other foreign finance mean 0.156; S.D. 0.364. Social development mean 0.218; S.D. 0.413. Soft power mean 0.106; S.D. 0.308. Stability mean 0.197; S.D. 0.398.
- Additional notes:
  - About 42 percent of estimates are published in peer-reviewed journals.
  - A quarter of estimates are from studies with at least one author affiliated with a Chinese institution.
  - Only 6 percent of estimates control for aid from other donors.
  - Chinese official finance is proxied by number of aid projects for 15 percent of the meta-dataset.

### IV. Conversion to common effect size and funnel analysis
- Conversion:
  - Partial correlation r from t-statistic t and degrees of freedom df: r = t / sqrt(t^2 + df).
  - Degrees of freedom approximated using the number of observations when df not reported.
- Funnel analysis:
  - Precision measured as inverse of the standard error of the partial correlation.
  - Weighted average partial correlation (inverse variance weights) = 0.003 (dashed vertical line in funnel plot).
  - Funnel plot appears broadly symmetric; prob-value for skewness is 0.15.
  - More precise estimates closely distributed around zero, suggesting potentially null genuine effects.

### V. Meta-regression analysis (FAT-PET) — specification and interpretation
- FAT-PET specification:
  - r_ij = β0 + β1 SE_ij + v_j + ε_ij, where r = partial correlation, SE = standard error of partial correlation, v_j = study fixed-effects.
  - Models estimated with WLS using precision squared (inverse variance) weights; standard errors clustered by study and recipient outcome.
- Interpretation:
  - FAT (β1 = 0) tests for funnel asymmetry/publication selection bias.
  - PET (β0 = 0) tests for genuine effect (precision-effect).

### VI. Meta-regression key results (selected)
- Baseline (All observations, 15 studies, 473 obs):
  - FAT (β1) = 0.205 (standard error 0.849)
  - PET (β0) = 0.002 (standard error 0.003)
  - RMSE 0.020
- By recipients’ outcome (Panel II):
  - Economic development (5 studies, 121 obs, 25.6%): FAT 1.371 *** (0.167); PET 0.005 *** (0.001); RMSE 0.006.
  - Social development (3 studies, 103 obs, 21.8%): FAT 1.187 *** (0.084); PET 0.003 *** (0.000); RMSE 0.014.
  - Governance (3 studies, 32 obs, 6.8%): FAT 8.615 *** (0.656); PET -0.071 *** (0.005); RMSE 0.008.
  - Stability (2 studies, 93 obs, 19.7%): FAT -0.018 (0.484); PET -0.017 (0.005); RMSE 0.051.
  - Other foreign finance (3 studies, 74 obs, 15.6%): FAT -2.603 (7.714); PET 0.017 (0.035); RMSE 0.047.
  - Soft power (2 studies, 50 obs, 10.6%): FAT 2.359 (0.870); PET -0.010 (0.003); RMSE 0.003.
- By publication outlet (Panel III):
  - Peer-reviewed journal (8 studies, 198 obs, 41.9%): FAT 1.145 *** (0.313); PET 0.006 *** (0.002); RMSE 0.031.
  - Working paper (7 studies, 275 obs, 58.1%): FAT -0.833 (1.489); PET 0.005 (0.006); RMSE 0.018.
- By author affiliation (Panel IV):
  - Chinese (2 studies, 117 obs, 24.7%): FAT -25.496 (14.057); PET 0.978 (0.534); RMSE 0.116.
  - Non-Chinese (13 studies, 356 obs, 75.3%): FAT 0.350 (0.809); PET 0.002 (0.003); RMSE 0.017.
- Interpretation highlights:
  - Baseline: no publication selection bias and no overall genuine effect.
  - Subsample heterogeneity:
    - Positive and statistically significant PET estimates for economic development and social development (but magnitudes negligible).
    - Governance shows a statistically significant negative PET estimate (very small in absolute terms).
    - No significant PET for stability, other foreign finance, or soft power.
  - Publication selection bias detected for peer-reviewed journal research.

### VII. Multiple meta-regression (heterogeneity analysis) — principal findings
- Specification:
  - r_ij = β0 + Σ βk Z_kij + β1 SE_ij + ε_ij, where Z are moderator variables; estimated without study fixed-effects.
- Main inferences (selected coefficients, general-to-specific where noted):
  - Baseline constant (col. 1) 0.011*** (0.002).
  - Recipients' outcome (relative to economic development omitted category):
    - Social development: general-to-specific -0.006*** (0.002).
    - Governance: general-to-specific -0.084*** (0.014).
    - Stability: general-to-specific -0.096*** (0.018).
    - Other foreign finance: general-to-specific -0.053*** (0.015).
    - Soft power: general-to-specific -0.082*** (0.015).
  - Chinese aid measurement:
    - ODA flows only: general-to-specific 0.069*** (0.014).
    - Dummy for Chinese projects: negative significant coefficients (e.g., -0.019***, -0.021**).
    - Number of Chinese projects: negative and significant in some columns (e.g., -0.024**).
  - Sample characteristics:
    - Macro data: negative coefficients in some models (e.g., -0.058*** in col. 7).
    - Region: Africa negative in some columns (e.g., -0.014***); LAC/Asia positive in some columns (e.g., 0.053*** in col. 7).
  - Publication outlet:
    - Peer-reviewed journal associated with positive and significant coefficients (e.g., 0.031***, 0.032***).
  - Affiliation:
    - Chinese affiliation coefficient not significant in reported model.
- Fit and sample:
  - RMSE around 0.019–0.021 across specifications.
  - Adjusted R^2 ranges from 0.126 to 0.278.
  - # studies 15; Observations 473.

### VIII. Robustness checks
- Exercises conducted:
  - Cluster standard errors by study only.
  - Use Fisher’s z-transform as alternative effect-size.
  - Expand meta-dataset by including interaction terms.
  - Exclude GMM estimates.
  - Exclude estimates on soft power outcomes.
  - Remove the study with largest number of reported estimates (Gehring et al. (2019)).
  - Leave-one-out checks excluding each of the 15 studies individually.
- Outcome:
  - Findings broadly remain quantitatively and qualitatively unchanged; detailed results in Appendix Tables A2 to A7.

### IX. Principal conclusions, limitations, and policy-relevant implications
- Principal conclusions:
  - No genuine overall empirical effect of Chinese foreign assistance on recipient countries in the full meta-dataset (473 estimates from 15 studies) after correcting for publication selection bias.
  - Heterogeneous subsample results:
    - Positive (but negligible) average effects on economic development and social development.
    - Negative (very small) effect on governance.
    - No statistically significant effects on socio-economic stability, inflows of other types of foreign finance, or citizens’ perceptions of China.
  - Publication selection bias present for peer-reviewed journal publications.
  - Study characteristics (dependent variable type, measurement of Chinese aid, geographic region, publication outlet) explain substantial variation in reported estimates.
- Comparative context:
  - Similarities with traditional aid literature on small or negligible average effects of aid on growth and mixed effects on governance and social outcomes.
  - Differences: patterns such as aid-associated conflict-fueling effects or increases in FDI from the same donor are not evident for Chinese official finance in this meta-dataset.
- Limitations:
  - Small number of papers underpin some sub-sample MRAs.
  - Excludes qualitative analyses, non-English studies, and methods not amenable to MRA (e.g., Computable General Equilibrium analysis).
- Suggested extensions:
  - Apply MRA to development effects of other Chinese flows such as trade and FDI.
  - Conduct MRA on the literature about determinants of Chinese aid allocation.
- Policy-relevant implication:
  - China’s pledge to develop a modern statistical information system for foreign assistance is identified as a welcome step toward transparency to enable further research.

*Has Chinese Aid Benefited Recipient Countries? Evidence from a Meta-Regression Analysis, Working Paper No. WP/22/46 — content derived from wpiea2022046-print-pdf*

### References .............................................................................................................

### wpiea2022046-print-pdf - References

### I. Introduction: context and motivation
- Since May 2020, the G20’s Debt Service Suspension Initiative (DSSI) provided eligible low-income countries a temporary moratorium on official bilateral debt repayments to free up resources for the COVID-19 pandemic response.
- China’s evolving role:
  - More than US$350 billion in official finance committed to developing countries between 2000 and 2014 (Dreher et al., 2017).
  - In Africa, Beijing is believed to have signed some 1,141 loan commitments worth US$153 billion over 2000-19.
  - China launched the US$1 trillion Belt and Road Initiative (BRI) in 2013.
- DSSI participation and Chinese share of G20 bilateral debt:
  - China’s share of bilateral debt owed by the world’s poorest countries to the G20 rose from 45 per cent in 2015 to 63 percent in 2019.
  - As of October 2020, China is the largest contributor to the DSSI, with over US$1.9 billion in repayments postponed, out of roughly US$5.3 billion suspended by G20 members for 44 debtor countries.
- Data transparency constraints:
  - China does not publish detailed country- and project-level foreign aid data; its aid is considered “a sensitive area [and] a state secret”.
  - China State Council’s White Paper (January 2021) reported RMB 270.2 billion (US$ 41.6 billion) in aid between 2013-18 but without granular country-year-sector breakdowns.
- AidData’s Global Chinese Official Finance Dataset:
  - Captures more than 4,000 Chinese government-financed projects in 138 countries over 2000-14.
  - Constructed using the Tracking Underreported Financial Flows (TUFF) methodology triangulating four source types: news reports in multiple languages; Chinese ministries/embassies/offices; counterpart countries’ aid and debt information systems; and case studies/field research.

### II. Research objective and approach
- Purpose of the paper:
  - Take stock of conflicting empirical literature on Chinese aid effectiveness.
  - Employ meta-regression analysis (MRA) using 473 estimates taken from 15 studies and obtained from regressions accounting for endogeneity.
- Why MRA:
  - Systematically review, summarize, and evaluate diverse empirical findings.
  - Identify and accommodate publication selection bias (preference for statistically significant results).
- MRA aims:
  - (i) examine presence of publication selection bias in Chinese aid effectiveness literature;
  - (ii) quantify the genuine effect of Beijing’s official finance on developing countries;
  - (iii) explain heterogeneity across reported results.

### III. Data compilation and scope
- Literature search and screening:
  - Initial search identified 190 studies; search ended in December 2020.
  - Screening excluded: (i) non-English papers; (ii) unpublished papers and theses (“grey” literature); (iii) papers unrelated to the research question; (iv) descriptive or qualitative studies without quantitative estimates.
- Meta-dataset construction follows Havránek et al. (2020) and MRA guidelines.
- Source of primary empirical estimates: studies that try to address identification problems and provide regression-based estimates.

### IV. Key empirical findings from the meta-regression
- Overall MRA results on the meta-dataset of 473 estimates:
  - Suggest absence of publication selection bias and absence of a genuine empirical effect on the full sample.
- Sub-sample MRAs by outcome category:
  - Positive impact of Beijing’s aid on economic and social outcomes in recipient countries.
  - Opposite effect on governance outcomes, albeit negligible in size.
  - No evidence that Beijing’s official assistance affected:
    - socio-economic stability in recipient countries;
    - inflows of aid from other donors (on average);
    - citizens’ perceptions of China (on average).
- Publication outlet effects:
  - Publication selection bias detected for research published in peer-reviewed journals (as opposed to working papers), suggesting studies failing to find statistically significant effects are less likely to be submitted or accepted by journals.
- Study-characteristic heterogeneity (factors explaining variation in estimates):
  - Effect sizes vary with:
    - type of development outcome considered;
    - how the Chinese aid variable is measured;
    - geographic region under study;
    - publication outlet.
  - Specific patterns:
    - Some evidence Chinese aid is more effective when it is concessional.
    - Chinese aid shows stronger effects when impact is assessed on economic outcomes.
    - Studies using samples of African countries or macro-level data report, on average, smaller aid-effectiveness coefficients.
- Authorship and institutional affiliation:
  - MRA results are not driven by authors’ institutional affiliation.

### V. Summary of the broader empirical literature (selective highlights)
- Economic outcomes:
  - Dreher et al. (2016, 2017) and Xu et al. (2019): Chinese projects boost economic growth in Africa (including confirmation using large developing-country samples).
  - Busse et al. (2016): no statistically significant and robust effect on economic growth in Africa.
  - Bluhm et al. (2020): Chinese infrastructure investments reduce within-region economic inequality in low- and middle-income countries; Xu et al. (2019) find the opposite for Africa.
- Social outcomes:
  - Martorano et al. (2020): areas with Chinese aid projects have better education and lower child mortality.
  - Cruzatti et al. (2020): mixed—infant mortality decreases at country level but increases at sub-national level.
  - Yuan (2020): beneficiary countries tend to score higher on the human development index.
- Governance and political effects:
  - Findings are mixed: evidence of local corruption increases (Brazys et al., 2017; Isaksson and Kotsadam, 2018a; Cha, 2020), undermining democratic governance, and disincentivizing economic reforms, while other studies find little evidence that Chinese aid helps autocrats stay in power or erodes citizens’ trust in government.
- Conflict:
  - Sardoschau and Jarotschkin (2019): Chinese aid projects in Africa associated with increased civilian riots at the district level.
  - Gehring et al. (2019): no conflict-fueling effect but correlation with more government repression and acceptance of authoritarian norms.
  - Strange et al. (2017): Chinese aid can substitute for sudden withdrawals of aid from traditional donors, potentially averting armed conflict.
- Environment:
  - Chen et al. (2020): public lending by China’s policy banks contributes to pollution due to financing carbon-intensive coal sector plants.
  - Marty et al. (2019): Chinese official finance reduced forest loss in Rwanda and Burundi.
  - Ben Yishay et al. (2016): deforestation linked to Chinese infrastructure projects only where domestic environmental enforcement is weak.
  - Hsiang and Sekar (2016): no evidence linking Chinese aid to increased illegal ivory production.
- Aid competition and complementarity:
  - Kilama (2016), Zeitz (2021), Hernandez (2017): evidence of reallocation or increased competitiveness from traditional donors in response to Chinese aid, often shifting toward infrastructure.
  - Humphrey and Michaelowa (2019): little change overall in multilateral and bilateral aid volumes in response to Chinese aid inflows.
- Soft power and perceptions:
  - Struver (2016), Raess et al. (2017): recipient countries tend to vote more similarly to China at the UN General Assembly.
  - Blair and Roessler (2018): African citizens near Chinese projects have more favorable perceptions of China; Xu et al. (2020) find infrastructure and social projects particularly influential.
  - Other studies report null or polarized effects (Sardoschau and Jaortschikin, 2019; Eichenauer et al., 2018).

### VI. Contribution and structure of the paper
- Threefold contribution:
  - First MRA of the empirical literature on Chinese aid effectiveness using a specifically compiled meta-dataset.
  - Identification of study characteristics explaining heterogeneity in reported estimates.
  - Comparison with the existing literature on traditional OECD DAC donors to address a research gap focused predominantly on Western donors.
- Paper organization:
  - Section II: overview of empirical literature on Chinese aid effectiveness.
  - Section III: data collection procedure underpinning MRA.
  - Section IV: MRA model, baseline results, and robustness checks.
  - Section V: exploration of study characteristics behind heterogeneity.
  - Section VI: conclusion.

*Italic: Content derived from wpiea2022046-print-pdf - References*

### Appendix Figure A1 provides a diagram illustrating the meta-dataset construction process.

### Appendix Figure A1 provides a diagram illustrating the meta-dataset construction process.

### Meta-dataset construction
- Final meta-dataset consists of 15 studies and 473 estimates, as referenced in Appendix Table A1.
- Included studies: those that investigated determinants of Chinese aid allocation and feature Chinese foreign assistance as an explanatory variable and a measure of recipient countries’ outcome as the dependent variable.
- Exclusions:
  - Regression coefficients from estimation techniques that do not account for endogeneity are excluded.
  - Two studies where Chinese aid solely entered as an interaction term were removed because coefficients are not directly comparable with those from linear models.
- When multiple versions exist (working paper and peer-reviewed, or several working paper versions), the most recent edition is retained to avoid double-counting.
- For each study the following are collected/coded:
  - sample size (number of observations), regression coefficients, and either standard errors or t-statistics;
  - paper title, publication year and outlet, author(s)’ name and affiliation;
  - type of development outcome (dependent variable);
  - measurement of the independent variable (Chinese aid);
  - sample of study and model specification.

### Dependent variable categories (recipients’ outcomes)
- Six organized categories:
  - (i) “economic development” — real GDP per capital growth, nighttime light intensity, Gini index for spatial concentration of activity;
  - (ii) “social development” — human development index score, household-level health and education outcomes (e.g., child/infant mortality, average years of education);
  - (iii) “governance” — perceptions and experience of corruption, citizens’ trust in government and engagement, democracy, government willingness to implement reforms;
  - (iv) “stability” — conflict, violence, social unrest;
  - (v) “other foreign finance” — development assistance from other donors and inflows of other types of finance from China, namely FDI;
  - (vi) “soft power” — attitudes towards China and political alignment with Beijing’s vote at the United Nations.

### Coding and descriptive statistics (Table 1 highlights)
- Sample size: N = 473 observations.
- Mean and S.D. for key coded variables (selected):
  - Partial correlation: Mean 0.001; S.D. 0.076.
  - SE (standard error of partial correlation): Mean 0.021; S.D. 0.016.
  - Author affiliation: Non-Chinese mean 0.753; S.D. 0.432. Chinese mean 0.247; S.D. 0.432.
  - Publication outlet: Working paper mean 0.581; S.D. 0.494. Peer-reviewed journal mean 0.419; S.D. 0.494.
  - Measurement of Chinese aid: All or non-ODA flows mean 0.552; S.D. 0.498. ODA flows only mean 0.448; S.D. 0.498.
  - Aggregate vs sector-specific: Sector-specific mean 0.201; S.D. 0.423. Aggregate mean 0.799; S.D. 0.423.
  - Variable form: Continuous mean 0.630; S.D. 0.483. Dummy mean 0.218; S.D. 0.413. Number of projects mean 0.152; S.D. 0.360.
  - Model specification: Without other donors mean 0.937; S.D. 0.244. Other donors mean 0.063; S.D. 0.244.
  - Sample level: Micro mean 0.638; S.D. 0.481. Macro mean 0.362; S.D. 0.481.
  - Region: Worldwide mean 0.307; S.D. 0.462. Africa mean 0.651; S.D. 0.477. Region: LAC/Asia mean 0.042; S.D. 0.201.
  - Recipients' outcome incidence (means):
    - Economic development mean 0.256; S.D. 0.437.
    - Governance mean 0.068; S.D. 0.251.
    - Other foreign finance mean 0.156; S.D. 0.364.
    - Social development mean 0.218; S.D. 0.413.
    - Soft power mean 0.106; S.D. 0.308.
    - Stability mean 0.197; S.D. 0.398.
- Additional descriptive notes:
  - About 42 percent of estimates are published in peer-reviewed journals.
  - A quarter of estimates are from studies with at least one author affiliated with a Chinese institution.
  - Majority have a regional focus, leverage micro-level data, and focus on economic and social outcomes.
  - Only 6 percent of estimates control for aid from other donors.
  - Chinese official finance is proxied by number of aid projects for 15 percent of the meta-dataset.
  - Most studies rely on an aggregate measure of aid and do not distinguish across flow types.

### Distribution of estimates and initial interpretation
- Of the 473 estimates:
  - 15 percent relate to negative and statistically significant effects of Chinese foreign assistance on recipient countries.
  - This is half the size of those reporting positive effects.
  - 55 percent find null effects.
- Initial interpretation: empirical literature in the meta-dataset lends little support to either strongly beneficial or harmful average effects of Beijing’s official finance but requires more rigorous methods to ensure “true” effects.

### Conversion to common effect size and funnel analysis
- Conversion formula used to obtain partial correlation r from t-statistic t and degrees of freedom df:
  - r = t / sqrt(t^2 + df)
- Degrees of freedom are approximated using the number of observations when df not reported.
- Partial correlations are unitless and allow comparability across differing dependent and explanatory measures.
- Funnel chart approach:
  - Precision is measured as the inverse of the standard error of the partial correlation.
  - In absence of publication selection bias, funnel plot should be symmetric.
  - Funnel plot (Figure 2) suggests potentially null genuine effects: more precise estimates closely distributed around zero.
  - Funnel plot appears broadly symmetric; prob-value for skewness is 0.15.
  - Dashed vertical line in funnel plot shows weighted average partial correlation 0.003 (using inverse variance weights).

### Meta-regression analysis (FAT-PET model and estimation)
- FAT-PET specification estimated:
  - r_ij = β0 + β1 SE_ij + v_j + ε_ij
  - r = partial correlation for ith estimate from jth study; SE = standard error of partial correlation.
  - Study fixed-effects v_j included; standard errors clustered by study and recipient outcome.
  - Model estimated with WLS using precision squared (inverse variance) as weights.
- Interpretation tests:
  - FAT (β1 = 0) tests for funnel asymmetry / publication selection bias.
  - PET (β0 = 0) tests for genuine effect (precision-effect).

### Meta-regression key results (Table 2 summary)
- Baseline (All observations, 15 studies, 473 obs):
  - FAT (β1) = 0.205 (standard error 0.849)
  - PET (β0) = 0.002 (standard error 0.003)
  - RMSE 0.020
- By recipients’ outcome (Panel II):
  - Economic development (5 studies, 121 obs, 25.6%): FAT 1.371 *** (0.167); PET 0.005 *** (0.001); RMSE 0.006.
  - Social development (3 studies, 103 obs, 21.8%): FAT 1.187 *** (0.084); PET 0.003 *** (0.000); RMSE 0.014.
  - Governance (3 studies, 32 obs, 6.8%): FAT 8.615 *** (0.656); PET -0.071 *** (0.005); RMSE 0.008.
  - Stability (2 studies, 93 obs, 19.7%): FAT -0.018 (0.484); PET -0.017 (0.005); RMSE 0.051.
  - Other foreign finance (3 studies, 74 obs, 15.6%): FAT -2.603 (7.714); PET 0.017 (0.035); RMSE 0.047.
  - Soft power (2 studies, 50 obs, 10.6%): FAT 2.359 (0.870); PET -0.010 (0.003); RMSE 0.003.
- By publication outlet (Panel III):
  - Peer-reviewed journal (8 studies, 198 obs, 41.9%): FAT 1.145 *** (0.313); PET 0.006 *** (0.002); RMSE 0.031.
  - Working paper (7 studies, 275 obs, 58.1%): FAT -0.833 (1.489); PET 0.005 (0.006); RMSE 0.018.
- By author affiliation (Panel IV):
  - Chinese (2 studies, 117 obs, 24.7%): FAT -25.496 (14.057); PET 0.978 (0.534); RMSE 0.116.
  - Non-Chinese (13 studies, 356 obs, 75.3%): FAT 0.350 (0.809); PET 0.002 (0.003); RMSE 0.017.
- Interpretation highlights:
  - Estimating over all 473 observations reveals no publication selection bias (baseline FAT not significant) and no overall genuine effect (baseline PET not significant).
  - Subsample MRAs reveal heterogeneity by outcome:
    - Positive but negligible genuine effects on economic development and social development (statistically significant PET in those subsamples).
    - Negative (but very small) effect on governance (statistically significant PET).
    - No significant effects for stability, other foreign finance, or soft power.
  - Substantial publication selection bias detected for research published in peer-reviewed journals (Panel III).
  - Institutional affiliation does not appear to systematically influence results (Panel IV).

### Multiple meta-regression (heterogeneity analysis) — key findings (Table 3)
- Model: r_ij = β0 + Σ βk Z_kij + β1 SE_ij + ε_ij, where Z are moderator variables; estimated without study fixed-effects.
- Main multiple MRA results and inferences:
  - Baseline constant (col. 1) 0.011*** (0.002).
  - Recipients' outcome moderators (relative to economic development omitted category):
    - Social development: negative coefficients in extended models; general-to-specific shows -0.006*** (0.002).
    - Governance: negative and statistically significant across specifications; general-to-specific shows -0.084*** (0.014).
    - Stability: negative and significant in many specifications; general-to-specific -0.096*** (0.018).
    - Other foreign finance: negative, general-to-specific -0.053*** (0.015).
    - Soft power: negative, general-to-specific -0.082*** (0.015).
  - Chinese aid measurement:
    - ODA flows only: positive and statistically significant coefficients; general-to-specific shows 0.069*** (0.014).
    - Dummy for Chinese projects: negative significant coefficients in several specifications (e.g., -0.019***, -0.021**).
    - Number of Chinese projects: negative and significant in earlier columns (e.g., -0.024**), but not robust across all columns.
  - Model specification:
    - Controlling for other donors does not show robust significant effect.
  - Sample characteristics:
    - Macro data: negative coefficients in some models (e.g., -0.058*** in col. 7).
    - Region: Africa shows negative in some columns (e.g., -0.014***), Latin America/Asia shows positive (e.g., 0.053*** in col. 7).
  - Affiliation and publication outlet:
    - Chinese affiliation coefficient not significant in reported model.
    - Peer-reviewed journal associated with positive and significant coefficients (e.g., 0.031***, 0.032***).
- Goodness of fit and sample:
  - RMSE around 0.019–0.021 across specifications.
  - Adjusted R^2 ranges from 0.126 to 0.278 across columns.
  - Number of studies 15; Observations 473.

### Robustness checks
- Robustness exercises performed:
  - Cluster standard errors by study only.
  - Use Fisher’s z-transform as alternative effect-size.
  - Expand meta-dataset by including interaction terms.
  - Exclude GMM estimates.
  - Exclude estimates on soft power outcomes.
  - Remove the study with largest number of reported estimates (Gehring et al. (2019)), which accounts for close to 15 percent of sample observations.
  - Leave-one-out checks excluding each of the 15 studies individually.
- Result: Findings broadly remain quantitatively and qualitatively unchanged; detailed results summarized in Appendix Tables A2 to A7.

### Principal conclusions and implications
- Overall meta-dataset (473 estimates from 15 studies):
  - No genuine overall empirical effect of Chinese foreign assistance on recipient countries after correcting for publication selection bias.
  - Subsample results indicate heterogeneous effects:
    - On average, positive (but negligible) effects on economic development and social development.
    - Negative (very small) effect on governance.
    - No statistically significant effects on socio-economic stability, inflows of other types of foreign finance, or citizens’ perceptions of China.
  - Publication selection bias detected when restricting to peer-reviewed journal publications.
  - Differences in study characteristics (dependent variable type, measurement of Chinese aid, geographic region, and publication outlet) explain substantial variation in reported estimates.
- Comparative context:
  - Results show similarities with traditional aid literature (e.g., small or negligible average effects of aid on growth; mixed effects on governance and social outcomes).
  - Differences from some Western aid literature where aid has been associated with conflict-fueling effects or increases in FDI from the same donor; such patterns are not evident for Chinese official finance in this meta-dataset.
- Limitations noted:
  - Small number of papers underpinning some sub-sample MRAs.
  - Meta-dataset excludes qualitative analyses, non-English studies, and studies using methods not amenable to MRA (e.g., Computable General Equilibrium analysis).
- Suggested extensions:
  - Apply MRA to investigate development effects of other Chinese flows such as trade and FDI.
  - Conduct MRA on the literature about determinants of Chinese aid allocation.
- Policy-relevant implication:
  - China’s recent pledge to develop a modern statistical information system for foreign assistance is identified as a welcome step towards transparency that could enable further research.

*Source: Appendix material and meta-regression results from the supplied PDF content.*

### REFERENCES

### REFERENCES

### Bibliographic scope and thematic coverage
- References cite studies on Chinese aid, development finance, aid effectiveness, and related empirical methods, including meta-analysis techniques and publication-bias tests.
- Geographic focus across cited works includes Africa, Latin America and the Caribbean (LAC), Asia, and worldwide comparative studies.
- Recurring topics in the literature: economic development, social development, governance, stability, soft power, other foreign finance, trade, FDI, environmental impacts, health outcomes, and debt/Belt and Road implications.
- Methodological references include meta-analysis, meta-regression, 2SLS, GMM, DiD, PPML, and guidelines for meta-analysis reporting.

### Representative cited works (authors and themes)
- Acker, Brautigam, Huang (2020): Debt Relief with Chinese Characteristics.
- Bluhm et al. (2020): Chinese infrastructure projects and economic diffusion.
- Brautigam (2009): The Dragon’s Gift: China in Africa.
- Dreher et al. (2016, 2017, 2021): Aid, China, and Growth; geography of China’s foreign assistance.
- Strange et al. (2013, 2017): China’s development finance data collection and aid–conflict nexus.
- Hurley, Morris, Portelance (2018): Debt implications of the Belt and Road Initiative.
- Empirical and methodological contributions: Doucouliagos & Paldam (2006, 2008, 2009, 2010), Stanley (2001, 2008), Stanley & Doucouliagos (2012), Egger et al. (1997), Cohen (1988).

---

### APPENDIX — Data and Meta-Regression Results

### Figure A1: PRISMA flow (study selection)
- Records identified through database searching: (n = 171)
- Additional records identified through other sources: (n = 19)
- Records after duplicates removed: (n = 147)
- Full-text articles assessed for eligibility: (n = 136)
- Full-text articles failing to meet inclusion criteria: (n = 121)
- Studies included in quantitative synthesis (meta-analysis): (n = 15)

### Table A1: Characteristics of Individual Studies (selected fields)
- 15 studies included in the meta-analysis. Examples of study entries:
  - Bluhm et al. (2020): Recipients' outcome(s): Economic dvpt; Sample: Africa; Asia; LAC; Worldwide; Study level: Micro; Measure of Chinese aid: Amount; dummy; nb. of projects; Type of Chinese aid flow: OF; Aggregate/Sectoral aid: Aggregate; sectoral; Estimation technique(s): 2SLS; % total reg.: 9.9%; Average partial corr.: 0.008 ***
  - Brazys and Vadlamannati (2020)*: Governance; Worldwide; Macro; Amount; ODA; OOF; OF; Aggregate; 2SLS; GMM; % total reg.: 3.8%; Average partial corr.: -0.094 ***
  - Busse et al. (2016)*: Economic dvpt; Africa; Macro; Amount; OF; Aggregate; GMM; % total reg.: 1.1%; Average partial corr.: 0.031 *
  - Dreher et al. (2017): Economic dvpt; Worldwide; Macro; Amount; nb. of projects; ODA; OOF & vague; Aggregate; sectoral; 2SLS; % total reg.: 3.8%; Average partial corr.: 0.037 ***
  - Martorano et al. (2020)*: Eco. dvpt; social dvpt; Africa; Micro; Dummy; OF; Aggregate; sectoral; DiD; % total reg.: 9.1%; Average partial corr.: 0.018 ***
  - Yuan (2020)*†: Social dvpt; Worldwide; Macro; Amount; OF; Aggregate; GMM; % total reg.: 11.0%; Average partial corr.: 0.021 ***
  - Notes: * published in peer-reviewed journal. † Chinese affiliation. OF: official finance. ODA: official development assistance. OOF: other official flow. OLS: ordinary least square. 2SLS: two-stage least square. GLS: generalized least square. GMM: general method of moments. DiD: difference-in-differences. PPML: Poisson pseudo-maximum likelihood. PCSE: panel corrected standard error. reg: regressions. corr. correlation. LAC: Latin America and the Caribbean. The sample average partial correlation is 0.003***.

### Table A2: FAT-PET Meta-Regression Results − One-way Cluster (selected entries)
- Baseline (All observations):
  - FAT (β1): 0.205 (0.985)
  - PET (β0): 0.002 (0.005)
  - RMSE: 0.020
  - # studies: 15
  - Obs.: 473
  - % obs.: 100%
- Recipients’ outcome — Economic development:
  - FAT (β1): 1.371 *** (0.167)
  - PET (β0): 0.005 *** (0.001)
  - RMSE: 0.006
  - # studies: 5
  - Obs.: 121
  - % obs.: 25.6%
- Recipients’ outcome — Governance:
  - FAT (β1): 8.615 *** (0.656)
  - PET (β0): -0.071 *** (0.005)
  - RMSE: 0.008
  - # studies: 3
  - Obs.: 32
  - % obs.: 6.8%
- Publication outlet — Peer-reviewed journal:
  - FAT (β1): 1.145** (0.345)
  - PET (β0): 0.006 ** (0.002)
  - RMSE: 0.031
  - # studies: 8
  - Obs.: 198
  - % obs.: 41.9%
- Author affiliation — Chinese:
  - FAT (β1): -25.496 (14.057)
  - PET (β0): 0.978 (0.534)
  - RMSE: 0.116
  - # studies: 2
  - Obs.: 117
  - % obs.: 24.7%
- Notes: Models estimated with WLS with precision squared weights and study fixed effects. Dependent variable: adjusted partial correlation between Chinese aid and recipients’ outcome. Brackets report standard errors adjusted for study and outcome level clustering. Cohen (1988)’s guidelines and Doucouliagos & Stanley (2013)’s guidelines provided. *p < 0.10, **p < 0.05, ***p < 0.01.

### Table A3: FAT-PET Meta-Regression Results − Fisher Partial Correlation (selected entries)
- Baseline (All observations):
  - FAT (β1): 0.153 (0.881)
  - PET (β0): 0.002 (0.004)
  - RMSE: 0.022
  - # studies: 15
  - Obs.: 473
  - % obs.: 100%
- Economic development:
  - FAT (β1): 1.371 *** (0.167)
  - PET (β0): 0.005 *** (0.001)
  - RMSE: 0.006
  - # studies: 5
  - Obs.: 121
  - % obs.: 25.6%
- Governance:
  - FAT (β1): 8.720 *** (0.618)
  - PET (β0): -0.072 *** (0.005)
  - RMSE: 0.008
  - # studies: 3
  - Obs.: 32
  - % obs.: 6.8%
- Peer-reviewed journal:
  - FAT (β1): 1.068** (0.419)
  - PET (β0): 0.007 *** (0.002)
  - RMSE: 0.031
  - # studies: 8
  - Obs.: 198
  - % obs.: 41.9%

### Table A4: FAT-PET Meta-Regression Results − Keeping Interactive Models (selected entries)
- Baseline (All observations):
  - FAT (β1): 0.362 (0.663)
  - PET (β0): 0.002 (0.003)
  - RMSE: 0.018
  - # studies: 15
  - Obs.: 636
  - % obs.: 100%
- Economic development:
  - FAT (β1): 1.143 *** (0.034)
  - PET (β0): 0.003 *** (0.000)
  - RMSE: 0.008
  - # studies: 5
  - Obs.: 189
  - % obs.: 29.7%
- Peer-reviewed journal:
  - FAT (β1): 1.068*** (0.220)
  - PET (β0): 0.006 ** (0.001)
  - RMSE: 0.031
  - # studies: 8
  - Obs.: 269
  - % obs.: 42.3%

### Table A5: FAT-PET Meta-Regression Results − Excluding GMM Estimates (selected entries)
- Baseline (All observations):
  - FAT (β1): 0.346 (0.812)
  - PET (β0): 0.002 (0.003)
  - RMSE: 0.017
  - # studies: 12
  - Obs.: 347
  - % obs.: 100%
- Economic development:
  - FAT (β1): 1.371 *** (0.172)
  - PET (β0): 0.005 ** (0.001)
  - RMSE: 0.006
  - # studies: 4
  - Obs.: 116
  - % obs.: 33.4%
- Peer-reviewed journal:
  - FAT (β1): 1.426*** (0.221)
  - PET (β0): 0.008 *** (0.001)
  - RMSE: 0.012
  - # studies: 5
  - Obs.: 72
  - % obs.: 20.7%

### Table A6: FAT-PET Meta-Regression Results − Excluding Soft Power (selected entries)
- Baseline (All observations):
  - FAT (β1): 0.460 *** (0.591)
  - PET (β0): 0.004 (0.003)
  - RMSE: 0.024
  - # studies: 14
  - Obs.: 423
  - % obs.: 100%
- Economic development:
  - FAT (β1): 1.371 *** (0.167)
  - PET (β0): 0.005 *** (0.001)
  - RMSE: 0.006
  - # studies: 5
  - Obs.: 121
  - % obs.: 28.6%
- Peer-reviewed journal:
  - FAT (β1): 1.144 *** (0.313)
  - PET (β0): 0.006 *** (0.002)
  - RMSE: 0.031
  - # studies: 8
  - Obs.: 198
  - % obs.: 46.8%

### Table A7: FAT-PET Meta-Regression Results − Removing Gehring et al. (2019) (selected entries)
- Baseline (All observations):
  - FAT (β1): 0.292 (0.977)
  - PET (β0): 0.002 (0.004)
  - RMSE: 0.021
  - # studies: 14
  - Obs.: 404
  - % obs.: 100%
- Economic development:
  - FAT (β1): 1.371 *** (0.167)
  - PET (β0): 0.005 *** (0.001)
  - RMSE: 0.006
  - # studies: 5
  - Obs.: 121
  - % obs.: 30.0%
- Peer-reviewed journal:
  - FAT (β1): 1.145*** (0.313)
  - PET (β0): 0.006 *** (0.002)
  - RMSE: 0.031
  - # studies: 8
  - Obs.: 198
  - % obs.: 49.0%

---

*Has Chinese Aid Benefited Recipient Countries? Evidence from a Meta-Regression Analysis, Working Paper No. WP/22/46*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022046-print-pdf.pdf_