| { | |
| "original_study": { | |
| "claim": { | |
| "hypothesis": "The quadratic association between violence and election fraud will be negative", | |
| "hypothesis_location": "p. 58 Violence and Fraud", | |
| "statement": "Fraud increases with violence up to a certain level, but then decreases again (coefficient on Violence (election, squared) term = -13.748, SE clustered at the regional command level = 4.720, p< 0.01)", | |
| "statement_location": "Table 2, Model 1, Violence (election, squared)", | |
| "study_type": "Observational" | |
| }, | |
| "data": { | |
| "source": "fraud data: computed from polling-station level data made public by the International Election Commission (IEC) on 19 September 2009 (The IEC publicly posted the data in three waves. The authors use the earliest data release - returns from 27,163 distinct polling stations on 19 September).\nviolence data: International Security Assistance Force (ISAF). These data, which are commonly known as ‘significant activity’ or SIGACT reports; Worldwide Incident Tracking System (WITS); Armed Conflict Location and Event Dataset (ACLED).\n development data: the 2007 National Risk and Vulnerability Assessment (NRVA) household survey, which was jointly administered by the Ministry for Rural Rehabilitation and Development (MRRD) and the Central Statistics Office (CSO) of Afghanistan.\ngeography data: US Geological Survey. GTOPO30 Digital Elevation Model and GIS spatial layers used for geocoding and aggregation.\n population data: LandScan (raster GIS datasets) the complete dataset is available at http://dvn.iq.harvard.edu/dvn/dv/nilsw", | |
| "wave_or_subset": "election data: 2009, development data: 2007, geography data: 2007, population: 2008.", | |
| "sample_size": "375", | |
| "unit_of_analysis": "district", | |
| "access_details": "The IEC publicly posted the data. WITS is available at http://wits.nctc.gov/. ACLED details are to be found in: Clionadh Raleigh et al. ‘Introducing ACLED: An Armed Conflict Location and Event Dataset’, Journal of Peace Research, 47 (2010), 651–60. US Geological Survey: Available at http://edc.usgs.gov/products/elevation/gtopo30/gtopo30.html. LandScan: Oak Ridge National Laboratory, LandScan Global Population database. http://dvn.iq.harvard.edu/dvn/dv/nilsw seems to be open access", | |
| "notes": "Unlike othe work on election fraud, the authors assume (and show) that insurgent violence is not biased in favour or against a particular candidate.This assumption characterizes violence around the Afghanistan election well, but it limits the applicability of the findings to other cases where this may not hold. Also, the authors limited their view of election fraud to manipulation tactics local to the polling centre. Moreover, lacking information about the precise location of polling stations, they were unable to estimate fraud at lower levels, for example, cities. Many of the covariates used in the regression analysis are available only at the district level. All violence indicators are reported as incidents per 1,000 population. The authors created two indicators for development: first, the proportion of households supplied with electricity, and, secondly, the per capita expenditure (in 1,000 s of afghanis)." | |
| }, | |
| "method": { | |
| "description": "The authors test the prediction that the relationship between violence and election fraud follows an inverted U-shape. Using district-level data from Afghanistan’s 2009 presidential election, the authors measure fraud through forensic last-digit tests and validate it with recount-based evidence.", | |
| "steps": "1. Collect all the data. \n2. Apply the last-digit test to the total vote count and code a binary dependent variable for fraud (which takes the value of 1 if this test is significant at the 5 per cent level for a particular district).\n3. Take violence data from ISAF SIGACT reports and calculate incidents per 1,000 population within five-day (20–24 August) and sixty-day windows around election day.\n4. Georeference ISAF SIGACT, WITS, and ACLED incidents to districts.\n5. Conduct robustness checks for the violence measure using alternative geo-referenced datasets (WITS and ACLED).\n6. Compute great-circle distances from district centres to Kabul.\n7. Derive average district elevation from GTOPO30 raster data.\n8. Aggregate population data from LandScan grid cells.\n9. Construct development indicators: proportion of households with electricity, and per capita expenditure.\n10. Add the number of closed polling stations at the district level as a control.\n11. Merge datasets.\n12. Estimate OLS models clustering standard errors at the regional command level.", | |
| "models": "logit regression", | |
| "outcome_variable": "Fraud, last-digit test (total count)", | |
| "independent_variables": "Violence (election), Violence (election, squared), Violence (2 months pre-election), Violence (2 months pre-election, squared), Electrification, Per-capita expenditure, Distance from Kabul, Elevation", | |
| "control_variables": "number of closed stations at the district level", | |
| "tools_software": "not stated" | |
| }, | |
| "results": { | |
| "summary": "Fraud initially rises with violence but then falls at higher levels, following an inverted U-shaped pattern. The coefficient on the Violence (election, squared) term is -13.748 (SE clustered at the regional command level = 4.720, p < 0.01).", | |
| "numerical_results": [ | |
| { | |
| "outcome_name": "Election Fraud", | |
| "value": "-13.748", | |
| "unit": "NA", | |
| "effect_size": "not stated", | |
| "confidence_interval": { | |
| "lower": "not stated", | |
| "upper": "not stated", | |
| "level": "not stated" | |
| }, | |
| "p_value": "<0.01", | |
| "statistical_significance": "true", | |
| "direction": "negative" | |
| } | |
| ] | |
| }, | |
| "metadata": { | |
| "original_paper_id": "10.1017/S0007123412000191", | |
| "original_paper_title": "Violence and Election Fraud: Evidence from Afghanistan.", | |
| "original_paper_code": "not stated", | |
| "original_paper_data": "http://dvn.iq.harvard.edu/dvn/dv/nilsw" | |
| } | |
| } | |
| } | |