## wpiea2021021-print-pdf - appendix Table A1). Droughts, although the only the seventh most common type of event

## Source details

**Canonical URL:** [wpiea2021021-print-pdf - appendix Table A1). Droughts, although the only the seventh most common type of event](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021021-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021021-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021021-print-pdf.pdf.json)

---

### Recent social unrest events: overview and COVID-19 period patterns
- Between January 2019 and January 2020, the events dataset identifies 59 unrest events in forty countries.
- Major protests late 2019–early 2020 occurred most notably in the Middle East and South America and were not directly linked to major natural disasters or epidemics.
- The 2019–2020 wave continued a longer trend since 2016, reversing a prior gradual decline following the 2011 Arab Spring peak.
- Since the start of the COVID-19 outbreak:
  - The number of major unrest events worldwide fell sharply and in March reached its lowest level in almost five years.
  - The decline in social unrest corresponds closely with a generalized decline in mobility driven by regulations (e.g., shelter-in-place orders) and voluntary social distancing.
  - Google Community Mobility Reports (retail and recreation; transit stations) show a remarkably close time-series association between reduced mobility and the decline in protests.
- Notable exceptions to the global pause in unrest include the United States and Lebanon; large protests there related to pre-existing issues (racial injustice in the United States and governance in Lebanon).
- Media coverage in the United States remained responsive: press articles related to unrest increased sharply concurrent with large street protests, suggesting media-based measures of unrest remain informative during a pandemic.

### Empirical methodology: goals and specifications
- Two main empirical findings reported:
  - A strong cross-sectional relationship: countries with more disasters also have more unrest, controlling for region, income, and exposure to waves of unrest.
  - A negative within-country relationship between epidemics and unrest, largest around 4-6 months after the epidemic starts; other disasters show no obvious within-country intertemporal relationship with unrest.
- Cross-sectional regression (equation (1)):
  - yi = α + Σβd xi,d + γr(i) + νy(i) + ei
  - yi = log number of social unrest events per capita since 1990 in country i.
  - xi,d = log number of disasters of type d.
  - γr(i) and νy(i) = fixed effects for region r(i) and income group y(i).
- Dynamic panel regression (equation (2)):
  - yit = αi + ηt + Σβj xit^j + γ′zit + eit
  - yit = indicator for social unrest event in country i in year t.
  - xit^j = indicator = 1 if the latest disaster occurred j months prior.
  - Linear probability model used to admit wide battery of fixed effects.
- Rationale for specification choices:
  - Local projection approach problematic because unrest and epidemics are rare (around 1.2 and 1.9 percent of country-months respectively), yielding low power.
  - Combining horizons (e.g., 4-6 months) increases power while preserving monthly frequency.
  - Comparison horizon uses more than 60 months after a disaster as baseline; coefficients measure conditional average unrest likelihood in the short run (less than 5 years) relative to 5+ years after a disaster.

### Cross-sectional evidence: main findings
- Figure 3 (descriptive) shows positive relationship across countries between per capita disasters and unrest for six disaster types; relationships stable across income groups.
- Appendix regression results (Table A2) confirm:
  - Positive and statistically significant cross-country relationship between disasters (all types combined) and social unrest (column (1)).
  - Relationships for other disaster types are statistically indistinguishable from each other (columns (2)-(4)) and robust across regions and income groups.
- Five-year within-country comparisons (with time fixed effects) similarly show robust and positive relationships; in many cases the relationship is slightly stronger, indicating countries with more disasters within a given five-year period also have more unrest.

### Dynamic (within-country) evidence: epidemics versus other disasters
- Key regularities from dynamic panel estimates:
  - Epidemics show a clear decline in the probability of unrest following an epidemic, peaking at 4-6 months after onset.
  - Other disasters (example shown: floods) show much weaker or no statistically significant dynamic patterns.
- Table 1: Impact of epidemics on unrest (linear probability model)
  - Baseline average frequency of unrest events (constant): 0.014***
  - Epidemic, current month coefficients:
    - Column (1): -0.011***
    - Column (2): -0.009*
    - Column (3): -0.008
    - Column (4): -0.011*
    - Column (5): -0.006
  - Epidemic, last 1-3 months coefficients:
    - Column (1): -0.009***
    - Column (2): -0.007
    - Column (3): -0.008
    - Column (4): -0.009
    - Column (5): -0.006
  - Epidemic, last 4-6 months coefficients (largest effect):
    - Column (1): -0.012***
    - Column (2): -0.011**
    - Column (3): -0.010*
    - Column (4): -0.010*
    - Column (5): -0.006
  - Epidemic, last 7-12 months coefficients:
    - Column (1): -0.006**
    - Columns (2)-(5): range from -0.001 to 0.003
  - Epidemic, last 13-24 months coefficients:
    - Column (1): -0.010***
    - Column (2): -0.008**
    - Column (3): -0.008*
    - Column (4): -0.010**
    - Column (5): -0.007
  - Epidemic, last 25-60 months coefficients:
    - Column (1): -0.006***
    - Columns (2)-(5): range -0.002 to -0.001
  - Months since last social unrest event: -0.0004*** (standard errors 0.00003–0.00004).
  - Months since last social unrest event, neighboring country: coefficients small (e.g., 0.00002 to 0.00004) and not consistently significant.
  - Deaths per capita in last epidemic: -0.0001*** and -0.0001** in shown specifications.
  - Normalized AIC by column: -1.437, -1.442, -1.056, -1.006, -1.127, -1.155.
  - R2 values by column: 0.014, 0.019, 0.05, 0.051, 0.049, 0.075.
  - Observations across columns: 27,505; 27,505; 18,123; 15,137; 12,953; 12,953.
  - Note: standard errors clustered at the country-month level.
- Table 2: Impact of flood on unrest (linear probability model)
  - Flood, current month coefficients:
    - Column (1): -0.007* (0.004)
    - Columns (2)-(5): range -0.003 to 0.003
  - Flood, last 1-3 months:
    - Column (1): -0.007* (0.004)
    - Other columns: small, not robust.
  - Flood, last 4-6 months: coefficients small and not statistically significant across specifications.
  - Flood, last 25-60 months:
    - Column (1): -0.006** (0.003)
    - Other columns: generally not significant.
  - Months since last social unrest event: -0.0004*** consistently.
  - Deaths per capita in last epidemic: -0.0001** in one specification; otherwise small.
  - Normalized AIC by column: -1.487, -1.49, -1.077, -1.013, -1.125, -1.15.
  - R2 values by column: 0.013, 0.017, 0.047, 0.051, 0.048, 0.072.
  - Observations across columns: 36,143; 36,143; 23,175; 19,033; 16,242; 16,242.
  - Note: standard errors clustered at the country-month level.

### Interpretation and identification considerations
- Cross-country positive relationship may reflect long-run scarring effects of disasters (epidemics included) or confounding by geography, institutions, or regional clustering.
- Within-country dynamic evidence:
  - Epidemics transiently reduce likelihood of unrest in the short run (peak suppression at 4-6 months).
  - Other disasters do not show similar sustained suppressive effects.
- Plausible mechanism:
  - Epidemics uniquely deter collective social activities due to contagion risk and health consequences; natural disasters hamper protest via transport/communications constraints but effects are more short-lived and limited to the activity itself.
- Identification caveats:
  - Natural disasters (e.g., earthquakes, storms) are plausibly close to random; epidemics may be related to country-specific conditions (e.g., health infrastructure).
  - Timing of epidemics at monthly level is argued to be plausibly random, supporting causal interpretation for tight windows.
  - Dynamic panel compares short-run windows to 5+ year baseline; if disasters have permanent effects this could bias estimates downward, but Appendix Figure A1 suggests such bias is quantitatively insignificant.

### Conclusion: synthesis of findings and implications
- The paper finds a positive cross-sectional relationship between epidemics (and more generally disasters) and social unrest, consistent with possible long-run scarring.
- In the short run, epidemics are associated with a reduction in the likelihood of social unrest, with the largest mitigating effect occurring around 4-6 months after epidemic onset.
- Recent COVID-19 patterns align with historic evidence: unrest was elevated before the crisis but declined during the pandemic due to mitigating factors; as the pandemic fades, unrest may reemerge where underlying social and political issues remain unresolved—not necessarily because of COVID-19 itself but because root causes persist.

### Appendix — Table A1: EM-DAT disasters since 1990 (with at least 50 observations)
- Flood: Number 4096; Avg Deaths 48; Avg affected 751467; Avg Damage (USD) 187812; Avg Mortality (%) 0.01
- Storm: Number 2942; Avg Deaths 139; Avg affected 321082; Avg Damage (USD) 483299; Avg Mortality (%) 0.04
- Epidemic: Number 1235; Avg Deaths 163; Avg affected 18993; Avg Damage (USD) 0; Avg Mortality (%) 0.86
- Earthquake: Number 819; Avg Deaths 1007; Avg affected 173415; Avg Damage (USD) 903272; Avg Mortality (%) 0.58
- Landslide: Number 523; Avg Deaths 50; Avg affected 13025; Avg Damage (USD) 15187; Avg Mortality (%) 0.39
- Extreme temperature: Number 522; Avg Deaths 335; Avg affected 198391; Avg Damage (USD) 105692; Avg Mortality (%) 0.17
- Drought: Number 472; Avg Deaths 51; Avg affected 3671603; Avg Damage (USD) 322349; Avg Mortality (%) 0.00
- Wildfire: Number 340; Avg Deaths 6; Avg affected 19438; Avg Damage (USD) 366170; Avg Mortality (%) 0.04
- Volcanic activity: Number 155; Avg Deaths 16; Avg affected 48054; Avg Damage (USD) 14949; Avg Mortality (%) 0.03

### Appendix — Table A2: Cross-section regressions (selected coefficients)
- Log number of disasters, per capita:
  - 0.515*** (0.034)
  - 0.793*** (0.028)
- Log number of droughts, per capita:
  - 0.434*** (0.063)
  - 0.411*** (0.057)
  - 0.418*** (0.058)
  - 0.626*** (0.037)
  - 0.626*** (0.037)
  - 0.624*** (0.037)
- Log number of epidemics, per capita:
  - 0.590*** (0.067)
  - 0.573*** (0.061)
  - 0.573*** (0.062)
  - 0.714*** (0.039)
  - 0.714*** (0.039)
  - 0.715*** (0.038)
- Regression metadata:
  - Regression type: x-sect (columns 1-4); 5-yr panel (columns 5-8)
  - Observations: 504 (cross-section), 795 (5-year panel)
  - R2 (examples): 0.281, 0.291, 0.363, 0.368, 0.503, 0.585, 0.585, 0.592

*Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021021-print-pdf.pdf*

### appendix Table A1). Droughts, although the only the seventh most common type of event

### wpiea2021021-print-pdf - appendix Table A1). Droughts, although the only the seventh most common type of event

### Recent social unrest events: overview and COVID-19 period patterns
- Between January 2019 and January 2020, the events dataset identifies 59 unrest events in forty countries.
- Major protests late 2019–early 2020 occurred most notably in the Middle East and South America and were not directly linked to major natural disasters or epidemics.
- The 2019–2020 wave continued a longer trend since 2016, reversing a prior gradual decline following the 2011 Arab Spring peak.
- Since the start of the COVID-19 outbreak:
  - The number of major unrest events worldwide fell sharply and in March reached its lowest level in almost five years.
  - The decline in social unrest corresponds closely with a generalized decline in mobility driven by regulations (e.g., shelter-in-place orders) and voluntary social distancing.
  - Google Community Mobility Reports (retail and recreation; transit stations) show a remarkably close time-series association between reduced mobility and the decline in protests.
- Notable exceptions to the global pause in unrest include the United States and Lebanon; large protests there related to pre-existing issues (racial injustice in the United States and governance in Lebanon).
- Media coverage in the United States remained responsive: press articles related to unrest increased sharply concurrent with large street protests, suggesting media-based measures of unrest remain informative during a pandemic.

### Empirical methodology: goals and specifications
- Two main empirical findings reported:
  - A strong cross-sectional relationship: countries with more disasters also have more unrest, controlling for region, income, and exposure to waves of unrest.
  - A negative within-country relationship between epidemics and unrest, largest around 4-6 months after the epidemic starts; other disasters show no obvious within-country intertemporal relationship with unrest.
- Cross-sectional regression (equation (1)):
  - yi = α + Σβd xi,d + γr(i) + νy(i) + ei
  - yi = log number of social unrest events per capita since 1990 in country i.
  - xi,d = log number of disasters of type d.
  - γr(i) and νy(i) = fixed effects for region r(i) and income group y(i).
- Dynamic panel regression (equation (2)):
  - yit = αi + ηt + Σβj xit^j + γ′zit + eit
  - yit = indicator for social unrest event in country i in year t.
  - xit^j = indicator = 1 if the latest disaster occurred j months prior.
  - Linear probability model used to admit wide battery of fixed effects.
- Rationale for specification choices:
  - Local projection approach problematic because unrest and epidemics are rare (around 1.2 and 1.9 percent of country-months respectively), yielding low power.
  - Combining horizons (e.g., 4-6 months) increases power while preserving monthly frequency.
  - Comparison horizon uses more than 60 months after a disaster as baseline; coefficients measure conditional average unrest likelihood in the short run (less than 5 years) relative to 5+ years after a disaster.

### Cross-sectional evidence: main findings
- Figure 3 (descriptive) shows positive relationship across countries between per capita disasters and unrest for six disaster types; relationships stable across income groups.
- Appendix regression results (Table A2) confirm:
  - Positive and statistically significant cross-country relationship between disasters (all types combined) and social unrest (column (1)).
  - Relationships for other disaster types are statistically indistinguishable from each other (columns (2)-(4)) and robust across regions and income groups.
- Five-year within-country comparisons (with time fixed effects) similarly show robust and positive relationships; in many cases the relationship is slightly stronger, indicating countries with more disasters within a given five-year period also have more unrest.

### Dynamic (within-country) evidence: epidemics versus other disasters
- Key regularities from dynamic panel estimates:
  - Epidemics show a clear decline in the probability of unrest following an epidemic, peaking at 4-6 months after onset.
  - Other disasters (example shown: floods) show much weaker or no statistically significant dynamic patterns.
- Table 1: Impact of epidemics on unrest (linear probability model)
  - Baseline average frequency of unrest events (constant): 0.014***
  - Country FEs: No in column (1); Yes in columns (2)-(6). Time FEs included only in column (6).
  - Epidemic, current month coefficients:
    - Column (1): -0.011***
    - Column (2): -0.009*
    - Column (3): -0.008
    - Column (4): -0.011*
    - Column (5): -0.006
  - Epidemic, last 1-3 months coefficients:
    - Column (1): -0.009***
    - Column (2): -0.007
    - Column (3): -0.008
    - Column (4): -0.009
    - Column (5): -0.006
  - Epidemic, last 4-6 months coefficients (largest effect):
    - Column (1): -0.012***
    - Column (2): -0.011**
    - Column (3): -0.010*
    - Column (4): -0.010*
    - Column (5): -0.006
  - Epidemic, last 7-12 months coefficients:
    - Column (1): -0.006**
    - Columns (2)-(5): range from -0.001 to 0.003 (statistical significance weak/absent)
  - Epidemic, last 13-24 months coefficients:
    - Column (1): -0.010***
    - Column (2): -0.008**
    - Column (3): -0.008*
    - Column (4): -0.010**
    - Column (5): -0.007
  - Epidemic, last 25-60 months coefficients:
    - Column (1): -0.006***
    - Columns (2)-(5): range -0.002 to -0.001 (less significant)
  - Months since last social unrest event: -0.0004*** (columns shown) with standard errors (0.00003)–(0.00004).
  - Months since last social unrest event, neighboring country: coefficients small (e.g., 0.00002 to 0.00004) and not consistently significant.
  - Deaths per capita in last epidemic: -0.0001*** (column shown) and -0.0001** (column shown), magnitudes very small.
  - Normalized AIC reported by column: -1.437, -1.442, -1.056, -1.006, -1.127, -1.155.
  - R2 values reported: 0.014, 0.019, 0.05, 0.051, 0.049, 0.075.
  - Observations across columns: 27,505; 27,505; 18,123; 15,137; 12,953; 12,953.
  - Note: standard errors clustered at the country-month level.
- Table 2: Impact of flood on unrest (linear probability model)
  - Flood, current month coefficients:
    - Column (1): -0.007* (0.004)
    - Columns (2)-(5): range -0.003 to 0.003 (not robust)
  - Flood, last 1-3 months:
    - Column (1): -0.007* (0.004)
    - Other columns: small, not robust.
  - Flood, last 4-6 months: coefficients small and not statistically significant across specifications.
  - Flood, last 25-60 months:
    - Column (1): -0.006** (0.003)
    - Other columns: generally not significant.
  - Months since last social unrest event: -0.0004*** consistently.
  - Deaths per capita in last epidemic: -0.0001** in one specification; otherwise small.
  - Normalized AIC reported by column: -1.487, -1.49, -1.077, -1.013, -1.125, -1.15.
  - R2 values reported: 0.013, 0.017, 0.047, 0.051, 0.048, 0.072.
  - Observations across columns: 36,143; 36,143; 23,175; 19,033; 16,242; 16,242.
  - Note: standard errors clustered at the country-month level.

### Interpretation and identification considerations
- Cross-country positive relationship may reflect long-run scarring effects of disasters (epidemics included) or confounding by geography, institutions, or regional clustering.
- Within-country dynamic evidence:
  - Epidemics transiently reduce likelihood of unrest in the short run (peak suppression at 4-6 months).
  - Other disasters do not show similar sustained suppressive effects.
- Plausible mechanism:
  - Epidemics uniquely deter collective social activities due to contagion risk and health consequences; natural disasters hamper protest via transport/communications constraints but effects are more short-lived and limited to the activity itself.
- Identification caveats:
  - Natural disasters (e.g., earthquakes, storms) are plausibly close to random; epidemics may be related to country-specific conditions (e.g., health infrastructure).
  - Timing of epidemics at monthly level is argued to be plausibly random, supporting causal interpretation for tight windows.
  - Dynamic panel compares short-run windows to 5+ year baseline; if disasters have permanent effects this could bias estimates downward, but Appendix Figure A1 suggests such bias is quantitatively insignificant.

### Conclusion: synthesis of findings and implications
- The paper finds a positive cross-sectional relationship between epidemics (and more generally disasters) and social unrest, consistent with possible long-run scarring.
- In the short run, epidemics are associated with a reduction in the likelihood of social unrest, with the largest mitigating effect occurring around 4-6 months after epidemic onset.
- Recent COVID-19 patterns align with historic evidence: unrest was elevated before the crisis but declined during the pandemic due to mitigating factors; as the pandemic fades, unrest may reemerge where underlying social and political issues remain unresolved—not necessarily because of COVID-19 itself but because root causes persist.

*Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021021-print-pdf.pdf*

### References

### References (wpiea2021021-print-pdf - References)

### Bibliographic references cited
- Aisen, Air and Francisco Veiga, 2013, “How Does Political Instability Affect Economic Growth?” European Journal of Political Economy, 29:151-167.
- Alesina, Alberto and Perotti, Roberto, 1996, “Income Distribution, Political Instability, and Investment”. European Economic Review, 40(6):1203-1228.
- Alesina, Alberto, Sule Özler, Nouriel Roubini, and Philip Swagel, 1996, “Political Instability and Economic Growth”, Journal of Economic Growth, 1:189-211.
- Barrett, Philip, Maximiliano Appendino, Kate Nguyen, and Jorge de Leon Miranda, 2020, Measuring Social Unrest Using Media Reports, IMF Working Paper 20/129.
- Barrett, Philip, Mariia Bondar, Sophia Chen, Mali Chivakul, Deniz Igan, forthcoming, Social Unrest and Financial Markets, IMF Working Paper.
- Bernal-Verdugo, Lorenzo E., Davide Furceri, and Dominique Guillaume, 2013, The Dynamic E_ect of Social and Political Instability on Output: The Role of Reforms. IMF Working Paper No. 13/91.
- Bristow, Nancy, 2017, American Pandemic: The Lost Worlds of the 1918 Influenza Epidemic, Oxford University Press.
- Cervellati, Matteo, Uwe Sunde, and Simona Valmori, 2017, “Pathogens, Weather Shocks and Civil Conflicts”, The Economic Journal, 127(607), Pages 2581–2616.
- Cervellati, Matteo, Elena Esposito, Uwe Sunde, and Simona Valmori, 2018, “Long-term Exposure to Malaria and Violence in Africa”, Economic Policy, 33(95), Pages 403–446.
- Collier, P., A. Hoeffler and D. Rohner, 2009, “Beyond Greed and Grievance: Feasibility and Civil War”, Oxford Economic Papers, No. 61, Pages 1–27.
- Deverell, William F., 2004, Whitewashed Adobe the Rise of Los Angeles and the Remaking of Its Mexican Past. University of California Press.
- Elledge, Jonn, 2020, “Revolts and Revolutions: How History Reveals the Ways Coronavirus Could Change Our World Forever”, Prospect, April 17, 2020.
- Fearon, J.D. and D.D. Laitin, 2003, “Ethnicity, Insurgency, and Civil War”, American Political Science Review, No. 97, Pages 75–90.
- Gonzales-Torres, Ada and Elena Esposito, 2017, Epidemics and Conflict: Evidence from the Ebola outbreak in Western Africa, mimeo.
- Hogarth, Rana A., 2017, Medicalizing Blackness Making Racial Difference in the Atlantic World, 1780-1840, The University of North Carolina Press.
- International Monetary Fund (IMF), 2020a, World Economic Outlook, Chapter 1: Global Prospects and Policies Global Prospects and Policies, Box 1.4.
- International Monetary Fund (IMF), 2020b, World Economic Outlook, Chapter 2: Dissecting the Economic Effects.
- Jong-A-Pin, Richard, 2009, “On the Measurement of Political Instability and Its Impact on Economic Growth”, European Journal of Political Economy, 25:15-29.
- Jordà, Òscar, 2005, “Estimation and Inference of Impulse Responses by Local Projections”, American Economic Review, 95(1):161-182.
- Miguel, Edward, Shanke Satyanath, and Ernest Sergenti, 2004, “Economic Shocks and Civil Conflict: An Instrumental Variables Approach”, Journal of Political Economy, 112(4), Pages 725-753.
- North, Douglass C. and Robert Paul, 1973, The Rise of the Western World: A New Economic History, Cambridge University Press.
- Ponticelli, Jacopo and Hans-Joachim Voth, 2020, “Austerity and Anarchy: Budget Cuts and Social Unrest in Europe, 1919-2008”, Journal of Comparative Economics 48 (1), Pages 1-19.
- Randall, David K., 2010, Black Death at the Golden Gate: The Race to Save America from the Bubonic Plague, W. W. Norton & Company.
- Snowden, Frank M., 2019, Epidemics and Society: From the Black Death to the Present, Yale University Press.

### Appendix — Table A1: EM-DAT disasters since 1990 (with at least 50 observations)
- Columns: Type | Number | Avg Deaths | Avg affected | Avg Damage (USD) | Avg Mortality (%)
- Flood: Number 4096; Avg Deaths 48; Avg affected 751467; Avg Damage (USD) 187812; Avg Mortality (%) 0.01
- Storm: Number 2942; Avg Deaths 139; Avg affected 321082; Avg Damage (USD) 483299; Avg Mortality (%) 0.04
- Epidemic: Number 1235; Avg Deaths 163; Avg affected 18993; Avg Damage (USD) 0; Avg Mortality (%) 0.86
- Earthquake: Number 819; Avg Deaths 1007; Avg affected 173415; Avg Damage (USD) 903272; Avg Mortality (%) 0.58
- Landslide: Number 523; Avg Deaths 50; Avg affected 13025; Avg Damage (USD) 15187; Avg Mortality (%) 0.39
- Extreme temperature: Number 522; Avg Deaths 335; Avg affected 198391; Avg Damage (USD) 105692; Avg Mortality (%) 0.17
- Drought: Number 472; Avg Deaths 51; Avg affected 3671603; Avg Damage (USD) 322349; Avg Mortality (%) 0.00
- Wildfire: Number 340; Avg Deaths 6; Avg affected 19438; Avg Damage (USD) 366170; Avg Mortality (%) 0.04
- Volcanic activity: Number 155; Avg Deaths 16; Avg affected 48054; Avg Damage (USD) 14949; Avg Mortality (%) 0.03

### Appendix — Table A2: Cross-section regressions (Dependent variable: Log number of social unrest events 1990-2019)
- Reported coefficients (with robust standard errors in parenthesis) and significance markers (***):
- Log number of disasters, per capita:
  - 0.515*** (0.034)
  - 0.793*** (0.028)
- Log number of droughts, per capita:
  - 0.434*** (0.063)
  - 0.411*** (0.057)
  - 0.418*** (0.058)
  - 0.626*** (0.037)
  - 0.626*** (0.037)
  - 0.624*** (0.037)
- Log number of earthquakes, per capita:
  - 0.466*** (0.063)
  - 0.454*** (0.058)
  - 0.457*** (0.058)
  - 0.745*** (0.041)
  - 0.745*** (0.041)
  - 0.748*** (0.041)
- Log number of epidemics, per capita:
  - 0.590*** (0.067)
  - 0.573*** (0.061)
  - 0.573*** (0.062)
  - 0.714*** (0.039)
  - 0.714*** (0.039)
  - 0.715*** (0.038)
- Log number of floods, per capita:
  - 0.723*** (0.147)
  - 0.758*** (0.137)
  - 0.743*** (0.136)
  - 0.968*** (0.048)
  - 0.968*** (0.048)
  - 0.967*** (0.048)
- Log number of landslides, per capita:
  - 0.499*** (0.071)
  - 0.480*** (0.065)
  - 0.481*** (0.065)
  - 0.771*** (0.040)
  - 0.771*** (0.040)
  - 0.773*** (0.040)
- Log number of storms, per capita:
  - 0.490*** (0.083)
  - 0.505*** (0.077)
  - 0.526*** (0.079)
  - 0.784*** (0.039)
  - 0.784*** (0.039)
  - 0.788*** (0.039)
- Constant terms (columns 1–8):
  - -0.974*** (0.066)
  - -1.003*** (0.070)
  - -0.983*** (0.087)
  - -1.291*** (0.202)
  - -0.942*** (0.067)
  - -0.962*** (0.084)
  - -0.962*** (0.084)
  - -0.935*** (0.136)
- Regression metadata:
  - Regression type: x-sect (columns 1-4); 5-yr panel (columns 5-8)
  - Region FEs: No, No, Yes, Yes, No, Yes, Yes, Yes
  - Time FEs: No, No, No, No, No, No, Yes, Yes
  - Income group FEs: No, No, No, Yes, No, No, No, Yes
  - Observations: 504, 504, 504, 504, 795, 795, 795, 795
  - R2: 0.281, 0.291, 0.363, 0.368, 0.503, 0.585, 0.585, 0.592
  - Adjusted R2: 0.280, 0.282, 0.350, 0.354, 0.503, 0.580, 0.580, 0.584
- Note: This table reports results of cross-sectional (columns 1-4) and 5-year panel (columns 5-8) regressions. The dependent variable is the log number of social unrest events over the 1990-2019 period. Robust standard errors shown in parenthesis.

### Appendix — Figure A1: Dynamic regressions versus local projections
- Note: Figure plots the point estimates and 90 percent confidence intervals of local projection and panel regression models. Source: Barrett et al. (2020), EM-DAT, and authors’ calculations.

*wpiea2021021-print-pdf - References*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021021-print-pdf.pdf_