This repository host results data and other information for the 2019 Afghan presidential election, as reported by the Independent Election Commission of Afghanistan (IEC). The repository also hosts historical data from past presidential elections, and administrative data (primarily polling center plans and voter registration data) released by the IEC in advance of the 2019 elections, which (after several previous postponements) were held on September 28 2019.
The primary output collects in tidy format all candidate results reported across all polling stations for which the IEC released data, and can be found in the prelim_af_candidate_ps_data_2019.csv and final_af_candidate_ps_data_2019.csv files.
These main polling station results output files are too large for Github (approximately 130 MB each) and are hosted remotely on Google Drive. A lightweight version of the results data, which includes only the polling station code, candidate code, and vote variables, is also available (for preliminary / final results). Accompanying keyfiles can be joined to expand these lightweight dataframes as needed.
This data is being released in open source for the purposes of contributing to public understanding of the elections and to allow for the analysis of available IEC reporting. Please read all accompanying documentation and link, cite, and credit as appropriate.
- Summary of files
- Summary of results data
- Background to the election
- Primary sources for this dataset
- Caveats, gaps, and errors
- Contact and acknowledgements
Each row in the primary polling station dataset (prelim / final) represents a candidate's vote total for each polling station. This data has also been re-aggregated into polling center (prelim / final)-, district (prelim / final)-, and provincial (prelim / final)-level summaries.
The main keyfiles collecting the metadata used to aggregate, disaggregate, and join the results datasets are:
- candidate_key.csv for a list of all registered candidates. Includes unique IEC-assigned ID codes, names in Dari (official IEC transliterations into English are pending until the results release), incumbency status, and gender.
- ps_key.csv for a list of all unique polling station codes, their corresponding parent polling center codes, designation for use by either male or female voters, and total number of assigned registered voters, as well as other metadata associated with reporting status and audit and invalidation decisions.
- pc_key.csv for a list of all planned (and non-planned) polling centers, their respective polling station counts, total voter registration data, and the center's reporting status in the 2018 parliamentary elections, among other metadata.
- district_key.csv for a list of all unique districts as reported by the IEC (as well as the administrative subdivisions of Kabul, Jalalabad, Mazar-e-Sharif, Kandahar, and Herat); see also the district_code_keyfile for a key used to track and match district codes across election cycles and datasets, drawing on the work of the Afghanistan District Maps project;
- province_key.csv for a list of provinces, their codes, and names in English, Dari, and Pashto.
A detailed variable codebook will be released in the near future. If you have questions about a variable definition or source, please feel free to contact the author. The underlying R code used to scrape, clean, and reorganize all data will also be published to allow for replication of results, although no guarantees are made as to its long-term functionality.
For a writeup of key findings from the election, see Colin Cookman, Assessing Afghanistan’s 2019 Presidential Election, U.S. Institute of Peace, August 2020, or the pre-proof copy of the report and accompanying source code in the analysis subfolder of this repository.
See also this preliminary analysis writeup (as of December 25 2019), or accompanying provincial or polling-station level summary reports.
A report on net changes between preliminary and final results per candidate per polling station can be found here; however, this does not correspond to the IEC's reported net change at the national level, as discussed further in the section below on apparent tabulation errors.
Afghanistan has conducted three nationwide presidential elections since the adoption of the 2004 Constitution; the 2014 election was the first in which President Hamid Karzai, constrained by term limits, did not return as an incumbent. After no candidate met the fifty percent majority requirement in the first round of that election, held in April 2014, former finance minister Ashraf Ghani and former foreign minister Abdullah Abdullah (previously the runner-up against Karzai in the 2009 elections) competed in a second round runoff election in June 2014. Abdullah and his supporters disputed the outcome of the second round, alleging fraud, and forced an UN-monitored audit of the results. The dispute was ultimately resolved through direct mediation by then-Secretary of State John Kerry that led to a power-sharing agreement between Ghani as president and Abdullah in a new "chief executive" position as part of a newly-formed 'National Unity Government'.
The unity government agreement included commitments to a process of electoral reform, which proceeded fitfully over the course of the government's first years in office, as multiple committees were formed and re-constituted to consider reform proposals. In September 2016 President Ghani issued a new electoral law by executive decree, overriding parliamentary objections, and in November 2016 new commissioners were appointed to the IEC and the ECC, which is responsible for adjudicating election complaints. The chairman of the IEC was again replaced a year later in November 2017. Amidst the ongoing electoral reform debates and IEC shakeups, parliamentary elections were indefinitely postponed past the five-year expiration of parliament's term in 2015, and after multiple delays were finally held in October 2018; see separate repository for results and analysis of those elections.
On December 30 2018, the IEC leadership announced the postponement of presidential elections from their original date of April 20 2019, to a new date of July 20 2019. In part in response to allegations of fraud and mismanagement of the parliamentary elections — the results of which took six months to finalize — on February 13 2019, President Ghani issued an executive order modifying Afghanistan's electoral law again and firing all members of the IEC and ECC. The eighteen already-registered presidential candidates nominated a pool of potential replacements and cast votes to elect the new commissioners; Ghani held final authority to decide whether an elected commissioner would be appointed to the IEC or ECC, however.
On March 20 2019, the presidential elections were further postponed by the new IEC leadership to a new date, September 28 2019. Elections for provincial councils, district councils (which have never been elected, despite being mandated under the 2004 Constitution), and parliamentary elections for Ghazni province (which were not held in 2018) were to have taken place simultaneous to the presidential election; however, on May 29 2019 IEC chairwoman Hawa Alam Nuristani announced that plans for those elections would be dropped, and did not specify a new date on which they might be held.
President Ghani's constitutional tenure expired on May 21 2019, but he remained in office (over the objection of his rivals) through the election period, with his extension endorsed by a Supreme Court ruling in April 2019. On August 6 2019, former foreign minister Zalmay Rasool withdrew from the race and endorsed Ghani. Following disputes within his coalition of supporters, former national security advisor Hanif Atmar suspended his campaign on August 8 2019; one of his running mates, Mohammad Mohaqiq, subsequently shifted his support to Abdullah Abdullah, but Atmar declined to endorse another campaign. On September 20 2019, candidate Shaida Mohammad Abdali also withdrew and endorsed Ghani.
Negotiations between U.S. diplomats and Taliban representatives in Qatar, and the possibility of a new power-sharing negotiation between Afghan factions as an outcome of those talks, overshadowed preparations for the elections, and public campaign activities by most of the candidates were limited for much of the official campaign period. President Ghani insisted that the elections proceed as scheduled and rejected alternative proposals for an interim government proposed by some of his rivals. The IEC began the distribution of election materials, including biometric voter verification devices, on August 30 2019, and announced that it had completed delivery of those materials in 33 out of 34 provinces as of September 16 2009. The official campaign period concluded on September 25 2019, and voting was held on September 28.
In the week after election day the IEC initially reported a provisional figure of approximately 2.69 million total votes cast, but later revised this to approximately 1.93 million biometrically verified votes, and later to 1.84 million after the removal of duplicate biometric vote signatures. Disputes over the process of auditing and recounting the initial vote in order to resolve discrepancies between polling station tally sheets and biometric voter records delayed the finalization of the preliminary results by several months; see section below on the audit and recount process. The IEC was initially scheduled to release preliminary results on October 19, but failed to do so; another deadline for the announcement of results on November 14 was also missed. Preliminary results were ultimately announced on December 22 2019, showing President Ghani with a narrow majority.
Following the announcement of preliminary results, provincial Electoral Complaints Commission offices took in approximately 16,500 complaints; on January 14 2020, the ECC said that 9,866 of these had been rejected, most for lack of supporting documentation, and that a recount had been ordered for 5316 polling stations. However, this order was not implemented as a further process of appeals against the provincial ECC decisions was undertaken by the national ECC. On February 5 2020, the national ECC issued a new set of orders calling for a sampling audit process and partial recount of three categories of disputed votes; see the section on the special audit process for further discussion. After a week and a half delay, the IEC began the process on Feb 16 2010 and shortly afterwards announced its completition and the release of final results on February 18 2010.
For additional background on the elections, see this Afghan Analysts Network primer, this FAQ by IFES, or the official IEC polling regulations. See also the author's summaries of contemporaneous news coverage here, Pajhwok elections coverage, and detailed analysis from the Afghan Analysts Network on IEC announcements of provisional results figures, disputes over biometrically verified voter data, the IEC's audit and recount decisions, and a summary of preliminary results data, among other items.
Formal candidate registration began on December 22 2018, and closed in early February 2019. On April 26 2019, the IEC assigned ballot positions, ballot symbols, and candidate ID codes to the eighteen registered presidential candidates, accepting all candidates who filed nominations to contest the elections. No candidates were assigned formal party affiliations on the ballot.
In a decision on November 6 2019, the IEC said that it would invalidate votes cast for withdrawn candidates Shaida Muhammad Abdali, Nur ul-Haq Ulumi, Haji Muhmmad Ibrahim Alekozai, and Zalmai Rassul, but not for Hanif Atmar, who the IEC said had not formally filed for withdrawal; the invalidated candidates' votes were not included in preliminary results reporting and will not be used for determining the national vote total required for a majority.
The preliminary list of registered candidates also includes vice presidential running mates; see also profiles of the tickets by the Afghan Analysts Network.
The IEC's preparations for the 2019 elections are, with some exceptions, based on the same universe of polling centers generated for the 2018 parliamentary elections. Out of an initial universe of 7366 polling centers that appeared on its website as of early June, in early August the IEC filtered that list to display only 5707 polling centers, without providing explanation for the change.
On August 19 2019, the IEC published a final polling center plan, which included details, including polling station counts and voter registration data, for 5373 polling centers. At a press conference on August 28 2019, Interior Minister Massoud Andarabi said that another 431 polling centers would be closed due to security threats, but did not specify which centers would be affected. Excluding centers in Ghazni province, where no parliamentary elections were held in 2018, 49 of the planned 2019 centers were not previously planned to open in 2018. 453 were planned to open in 2018 but were not included (either due to disqualification or failure to report results, IEC reporting has not clarified either way) in the 2018 preliminary results, and 506 were not included in final results.
In total, 4540 polling centers and 25079 polling stations were included in the publication of preliminary results on December 22, although this includes some (but not all) stations reporting zero votes, and omits some stations that opened but were invalidated in audit decisions. 4540 polling centers were also included in the final resuls, and 25075 polling stations (eight stations included in the preliminary results were omitted in the final results, and four stations omitted in the preliminary results were included in the final results). The IEC released a list of 3006 closed polling stations on December 28; see the polling station status report file, with caveats in the section on gaps and discrepancies below.
The IEC planned for 29,586 polling stations at the 5373 polling centers — 18,467 stations assigned for use by male voters, and 11,119 assigned for female voters. (Each polling station represents a booth within the polling center, from a minimum of one station per center to a maximum of 36.) The IEC did not publish a pre-election polling station plan for the 2018 parliamentary elections that classified stations for male or female voters, which would allow for direct comparison, but in those elections reported preliminary results from 17,546 polling stations, and final results from 17,713 stations. The increase in the number of planned stations is in part due to a decision by the IEC for 2019 to lower the maximum ballot allocation / registration per station to 400. In Ghazni, 828 polling stations are planned that would have been ommited from 2018 plans.
Polling center and polling station codes do not correspond across election periods (nor do provincial or district codes - see the district_code_keyfile for a join key at those levels). Using a combination of methods including geocoordinate matches (using latitude and longitude coordinates for 2018 centers, and the Afghanistan Election Data project's reported coordinates for 2014 centers), matches by center name and district, and matches by rough geocoordinate proximity, the author was able to link 5,330 of 6,792 polling centers available from 2014 to corresponding sites planned in 2018. The pc_code_matching_2014_2018.csv file details the relationships between 2014 and 2018 sites as determined through this process, and the file nearest_replacement_center_2014_2018.csv identifies nearest replacement centers for 2014 sites that appear to have been relocated. Note that the geocoordinate data sources available for this matching process in many cases did not align exactly between the two periods, even when they appear to specify the location of the same named building. The process of manual matching almost certainly introduced some further errors or omissions, meaning caution should be advised in drawing firm conclusions from this dataset.
Afghanistan introduced polling center-based voter registration lists for the first time in the 2018 parliamentary elections, and that data provides a baseline for 2019 voter registration. In June 2019, the IEC conducted an approximately monthlong "top-up" registration exercise to register newly eligible voters or otherwise update voter rolls; registration changes were only possible to be made at a limited number of polling centers, listed by the IEC here.
On the week of July 1 2019, the IEC announced preliminary figures from this top-up exercise, and published provincial-level reports for a total of 5056 polling centers (identified by name only and not polling center code). Ghazni province was ommitted from these provincial reports. The preliminary data ommited 15 polling centers with previously known voter registration data, and added registration data for five polling centers that had no previously recorded registration data (or votes) in 2018.
In an accompanying statement to the July update, the IEC stated that it had added a net increase of +317,395 new voters through the top-up exercise, although it emphasized that the data released at this time was not final. It also said that registration would continue past the one month top-up period in Ghazni. The actual net change from the detailed preliminary provincial reports released alongside this statement was a decline of -8443 registered voters from 2018, however; the reported preliminary 2019 registration (minus Ghazni) totaled 8.834 million, compared to 8.843 million total known registered voters in 2018. In subsequent public comments, IEC officials were quoted saying that another approximately 400,000 "ghost voters" or duplicate registrations had also been removed from the rolls, but no explanation was offered for the earlier discrepancies or details provided on these adjustments.
As detailed in the polling center key, the August 19 publication of the IEC's pre-election polling center plan included final voter registration data for all planned centers, superseding the earlier preliminary registration data. These new final registration figures for 2019 total 9.665 million voters for all planned polling centers. Excluding Ghazni, for which there was no previous 2018 registration data, this represents a net increase of approximately +595,000 voters over the earlier preliminary 2019 figures, and approximately +587,000 over 2018 data. (The new total voter registration for Ghazni is 235,213 registered voters.)
Pajhwok subsequently reported that the 2018 voter rolls were updated in early 2019 to add "[voter registration] books that had not been scanned and some pages that had been missing", prior to the removal of -428,901 duplicate or 'ghost' voter records, which produced a new total of 9.103 million voters. Though not specified in the Pajhwok report, by inference this would indicate approximately +689,000 registrations were added prior to the duplicate record removals, for a net increase of approximately +260,000 total registered voters over 2018 prior to the top-exercise conducted in June. This would suggest a net increase of approximately +562,000 registered voters during the top-up exercise, or around +327,000 additional voters over 2018 if the addition of Ghazni is discounted. The IEC has again not offered a public explanation for the discrepanacy between these figures and earlier statements.
The 2019 elections are first in which the IEC has specified polling station-level voter lists; under current planned procedures, voters should undergo a biometric verification process prior to casting their vote at their assigned station. As noted above, the IEC has lowered the maximum number of ballots allocated to a polling station from 600 in previous presidential elections to 400, increasing the number of stations and accompanying staff in some centers. The polling station key includes the number of registered voters assigned to each polling station as reported by the IEC in its pre-election polling station plan.
Election results were transmitted to the IEC through at least two separate reporting streams: physical tally sheets prepared by polling station workers and transmitted to the National Tally Center in Kabul, and digital records from biometric voter verifiction (BVV) devices provided by the German company Dermalog, which polling staff were instructed to use to check in voters, to generate unique QR code stickers to affix to ballots after votes were cast, and to transmit a summary report of the results sheet as well. (Digital scans of the physical tally sheets were also transmitted through the BVV devices, and made available to candidate teams for preview on a restricted-access server.)
Discrepancies between these reporting streams quickly emerged. Ultimately, the IEC affirmed that it would only accept votes where the voter had been biometrically verified as valid. However, because the biometric data did not include candidate choice, any polling station where there was a discrepancy between the vote totals on the physical tally sheet (which did record individual candidate totals) and the BVV device (which did not) would have to undergo a recount or audit in order to identify ballots with verified QR code stickers in order to produce an updated and reconciled candidate vote total.
The IEC initially released five main decisions regarding disputes, audits, and recount orders:
- Decision 104 of November 4 2019 — covering the selection of 5439 polling stations for audit for which biometric data was available (partially available in English or in full in Dari);
- Decision 105 of November 7 2019 — covering the selection of 2423 polling stations for audit for which biometric data was missing but which were believed to have been open or for which there were physical tally sheet results (in English / Dari);
- Decision 108 of November 14 2019 — detailing 137630 biometrically verified votes at 4563 polling stations which were initially quarantined by Dermalog for a variety of criteria but which the IEC ultimately accepted for inclusion (in English / Dari);
- Decision 109 of November 15 2019 — detailing 102012 biometrically verified votes at 7354 stations with accompanying timestamp data idicating votes were recorded outside of official voting hours, the majority of which the IEC also accepted, but 262 of which were ordered for audit (in English / Dari); and
- Decision 112 of November 21 2019 — detailing 22 additional polling stations (although the IEC reported a total count of 23 stations) that had been referred for inquiry by National Tally Center staff, of which 13 were ordered for audit (in Dari only).
On November 28 2019, the IEC released a detailed list of the stations it had selected for audit or recount in each province, based on twelve different criteria identified under Decisions 104 and 105 (including some referrals from the ECC, of which 393 were not otherwise covered by other IEC audit criteria). For Decision 105, 1287 stations under investigation had physical tally sheet results data reported through the National Tally Center but no accompanying biometric data, and 1136 stations were believed by the IEC to have been open on election day but no results data of any form was available. For stations covered under Decision 104, total biometric votes per station were also reported in the IEC's audit list. The specific 262 stations identified for audit under Decision 109 were not made public. Decision 112 included the specific polling station codes of the affected stations but no other details.
The Abdullah campaign and several other candidates objected to the IEC's decisions and opposed the recount and audit process, blocking it from being completed in seven provinces (Badakhshan, Baghlan, Faryab, Jawzjan, Panjshir, Samangan, and Takhar) for several weeks before ultimately relenting and allowing the audit to be completed. The three main categories of complaints from the Abdullah team include the IEC decision to accept approximately approximately 101,000 outside of hours votes (although the IEC ordered audits for 262 polling stations in Decision 109); the IEC decision to accept approximately 137,000 votes at previously quarantined polling stations (under Decision 108); and the decision to investigate and audit (rather than automatically invalidate) polling stations that were missing biometric vote records (under Decision 105). At the time of its initial complaint, the Abdullah team appears to have estimated approximately 70,000 votes at 724 polling stations fell into the third category, although the IEC and ECC ultimately identified 2423 stations in total that had these missing biometric data problems.
The polling station audit target list file lists all publicly identified stations ordered for audit and recount under Decisions 104, 105, and 112. Separate author correspondence with election observers indicates that in total, 8368 unique polling stations were selected for audit or recount under these or other criteria (including referrals from the Electoral Complaints Commission), but the IEC's published audit target list does not reflect this, and a complete list of those stations is currently unavailable.
On December 28 2019, the IEC released a list of invalidated polling stations based on decisions made by the IEC and ECC against the original audit list, and updated this again the following day to also specify some closed polling stations, for a total of 5305 polling stations (3006 closed, 2299 invalidated) listed in total across all provinces. Invalidation decisions were based on IEC Decision 113, Decision 116, or Decision 117 (released in Dari only, in conjunction with the announcement of the preliminary results on December 22), or by a decision of the ECC.
727 of these invalidated stations were also included in the preliminary results, all of which were zero vote counts. (Another 457 stations not on the invalidated list also reported zero votes in the preliminary results.) This list also included reported results sheet vote totals prior to invalidation for 760 stations (102,021 votes in total). This information is detailed in the invalidated polling station list file, and was manually transcribed in the invalidated_ps_vote_tallies.xlsx file.
There are some discrepancies and gaps in the available reporting by the IEC; in total, 10 previously planned polling stations, all of which were previously selected for audit, did not report any preliminary results but were also not included in invalidated or closed polling station list. See section below on gaps in the available data for more discussion of these and other disputed votes categories.
As described above, following the announcement of preliminary results, the provincial offices of the Electoral Complaints Commission received approximately 16,500 complaints filed by the competing candidates. In its announcement of January 14 2020, the ECC said that a recount had been ordered for 5316 polling stations, but a detailed list of those stations was not released at the time. A further period of appeals to the national ECC against the decisions made by the provincial ECCs led to a subsequent announcement on February 5 2020, in which the national ECC issued a new set of orders calling for a sampling process and partial audit and recount of three main categories of disputed votes raised by the Abdullah campaign.
- Out-of-hours votes: The ECC ordered a special audit for 1103 polling stations out of 7354 polling stations that reported a total of 102012 votes recorded outside of regular voting hours (a 15% station sample). If at least 35% of the sample (386 stations) were found to not have biometric information, a polling station journal, and a valid election day results sheet, all 102012 votes at all 7354 polling stations would be invalidated.
- Initially quarantined stations: The ECC ordered a special audit for 309 polling stations out of 3097 stations (a 10% station sample) that were initially flagged for quarantine due to discrepancies between biometric voter records that suggested "extra" votes. (During the initial quarantine process, there were also 1466 polling stations where discrepancies suggested "fewer" votes - these stations do not appear to have been subject to sampling or audit. See note below for further explanation of the quarantined category of disputed votes.) If at least 35% of the sample (108 stations) were found to not have the aforementioned information, all polling stations in this category would go to audit. Any polling stations that met the criteria would be validated, and any polling stations that failed to meet the criteria would be fully invalidated.
- Missing biometric records: The ECC ordered a recount for 298 polling stations out of 2423 previously flagged as having been missing biometric voter data. 291 of these stations were all previously included in the preliminary results, although 37 reported zero votes. Seven of these stations were ommited from the preliminary results.
On February 15 2020 the IEC published a list of polling stations selected for sample or recount on the basis of these orders, which is summarized in the iec_ecc_special_audit_targets.csv file. There are discrepancies between the stations listed and the ECC audit orders; only 1085 stations are listed for the out-of-hours sample (18 fewer than ordered), whereas 311 stations are listed for the initially quarantined stations (two more than ordered).
As of this writing, the IEC has not provided any detailed reporting on the outcome of this audit process beyond the announcement of final results.
As part of the 2018 parliamentary elections results project, the author received a dataset of latitude / longitude coordinates for 7413 known polling centers from an election observer. Because the 2019 election polling center plan is based on the same list as 2018, this location data also corresponds to the 2019 centers.
Although polling centers are public sites, these coordinates were shared with the request that they not be published in raw format so as to avoid any potential security risk to the physical polling center locations. Accordingly, absent reporting of this information by another source, the raw coordinate data will not be included in this dataset.
However, the available coordinate data was used to calculate nearest neighbor distances for the five closest polling centers for each respective center (in kilometers using the Vincenty ellipsoid formula, and in lat/lon decimal degrees), which can be found in the 2018 repository's pc_plan_nearest_neighbor_coordinates.csv file and may be used for further geospatial analysis of results. (See also note below on ongoing efforts to match polling centers to previously published 2014 geocoordinates.)
As part of its preliminary results reporting, the IEC released scanned images of post-audit polling station results forms, which were downloaded as part of the data scrape process and may potentially be used as a visual check on the machine-readable results. Those images are too large to host in Github, and are mirrored on Google Drive here (together with scans listed separately by the IEC in its report on invalidated polling stations).
For a list of all available polling station scan links, see the files ps_scan_links and invalidated_ps_scan_links. In some cases, links were not working at the time of download, producing 404 load errors; for lists of these broken links, see the ps_scans_broken_links invalidated_ps_scan_broken_links files.
Not all polling stations were accompanied by a scanned polling station results form. For a provincial-level summary highlighting gaps between reporting polling stations and scans missing from the IEC results site, see the ps_scans_reporting_check file, or otherwise the polling station status report for specific stations missing results scans.
This dataset also hosts cleaned and restructured preliminary and final certified results data from Afghanistan's previous three presidential elections held in 2004, 2009, and 2014 (both first round and second round runoff). The full polling station-level results files for the 2009 elections are too large to be hosted on Github but are hosted remotely on Google Drive here (prelim / final); the ps_lite files in each subfolder includes the basic polling station, candidate code, and vote data.
In addition to past presidential election cycles in this repository, the author's dataset of Afghan parliamentary elections results (for the 2005, 2010, and 2018 parliamentary elections) can be found in a separate repository.
The historical presidential election results data were drawn wherever available from the original IEC results, as found on its website during the June-September 2019 period (see results sections for 2004, 2009, and 2014). (For the 2004 and 2005 election cycles, elections were administered by the Joint Election Monitoring Body, a joint UN-Afghan government body.) The IEC's results reporting sections offer varying levels of detail or completeness; only provincial-level data is available for the 2004 elections. Some older sections of the IEC website have also been broken by subsequent website redesigns, making it currently impossible to cross-check more detailed polling center-level results on the web against pdf releases. Use of the Wayback Machine is recommended as an additional reference for contemporaneous presentations of the results, dating back to 2008.
Some historical election data was also drawn from or otherwise cross-referenced with the Afghanistan Election Data project (a project of the National Democratic Institute and Development Seed). (Although note that the AED project only offers preliminary results for the 2009 and 2014 elections.) For this repository, detailed results for the 2014 run-off election were also derived in part from a internal IEC document summarizing the outcome of the 2014 second round audit, which was obtained from an election observer. (As part of the national unity government agreement, the final results of the 2014 elections were initially withheld from publication, but were later released by the IEC in February 2016.)
The underlying R code used to clean up the available historical results data, and the raw IEC files used for this purpose, are included their respective subfolders. Some additional manual cleaning of older data was also undertaken that may not be fully reflected in this code, and caution in advised in running it without careful checks on the process.
For mapping purposes, UN OCHA has published a set of shapefiles for district boundaries, which can be found on the Humanitarian Data Exchange and is also included in the district data subfolder of the 2018 parliamentary elections project for reference. Further transformations are required to re-base 2018-19 districts or election results data to match these district boundaries, however. For an unofficial, ArcGIS-only version of the 421 districts used in the most recent IEC data, see this Map Sync map.
UN OCHA published metadata for updated district boundaries on October 22 2019, but has restricted access to the shapefiles, referring users to purchase the data through the Afghan Geodesy and Cartography Head Office (AGCHO).
This repository remains a work in progress and caution is advised when using it for analysis. The author cannot verify the accuracy of, or account for any discrepancies in, the underlying data, and makes no guarantees as to its completeness. The results collected may be further altered by legal complaints, IEC amendments to the results, or other disputes and adjudication processes.
Known current gaps in the available data include:
While the IEC announcement of final results shows a small net change of -453 votes between final and preliminary results, this is not reflected in the more detailed polling-station level data released by the IEC, which only shows a net change of -10 votes. Net changes by candidate per polling station are reported in the prelim_final_net_change.csv file file. This difference between preliminary and final vote totals is the result of the ommission of eight polling stations (0904097-03, 0904097-04, 0904097-05, 0904097-06, 0904097-07, 0904097-08, 0913265-08, and 0913265-09, all in Baghlan province) that were previously included in the preliminary results from the final results, although only two of these stations (0913265-08 and 0913265-09) had previously reported any votes.
None of the omitted stations were among those selected for special audit by the ECC, so the reason for their omission is unclear in the absence of any further explanation from the IEC. Another four stations (0616369-03 in Nangarhar and 2711172-10, 2711172-11, and 2711172-12 in Kandahar) were included in the final results after being selected for ECC audit and having previously been omitted from the preliminary results, although all four of these stations reported zero votes in the final results.
As noted above, the IEC has not released any detailed reporting on the outcome of the special audit process conducted between February 18-20 in response to ECC orders, beyond announcing the final results data. The ECC also did not release a detailed list of polling stations covered under the audit and recount orders made by the provincial ECC offices, initially announced on January 14 2020, and subsequently over-ridden by the national ECC order of February 5 2020. An election observer source provided an internal IEC report to election stakeholders that indicated that 23 polling stations had been removed as a result of the audit, but that report did not specify the criteria under which this had been done; of these stations, 14 still appear in the published results. (The 23 reportedly invalidated stations are: 0104420-04 and 0115544-06 in Kabul; 0608218-05, 0616365-06, 0604106-04, and 0621431-06 in Nangarhar; 0913265-09, 0904097-08, 0913265-08, 0904097-04, and 0913265-02 in Baghlan; 1110176-05 and 1113279-07 in Ghazni; 2701018-13, 2701039-25, 2704077-01, 2711169-03, and 2713189-02 in Kandahar; 2908169-12, 2908169-17, 2907131-02, and 2907135-01 in Faryab; and 3001024-15 in Helmand.) Had all 23 been removed, this still would not have produced the net changes in the published IEC national-level results.
Additional gaps in reporting on the audit decisions includes:
- Duplicate votes: On November 2 2019 the IEC announced that Dermalog had identified 86225 duplicate biometric votes through a process of comparing digital biometric signatures, providing a new total of approximately 1.84 million total verified votes (prior to audit and recount). No detailed breakdown of the duplicated votes or the polling stations at which they were recorded has been released; the Afghan Analysts Network has reported a provincial breakdown. The IEC did not publically clarify a procedure for removing these duplicate votes, but an election observer source told this author that all of the duplicate votes fell within polling stations already selected for audit under other criteria, and would be removed as part of the audit and recount process there; this account is unverified, however.
- Out of hours votes: No detailed information has been released by the IEC on the full list of stations (covered under IEC Decision 109 and the ECC special audit order) where out of hours votes were reported, beyond those later selected for sample audit in response to the ECC order. An election observer source reports those stations selected for audit under Decision 109 (reportedly comprising 13949 votes out of approximately 101,000 in total) fell into one of these three categories: votes were recorded late at night after polls closed following a normal day's operation; votes were recorded early in the morning prior to opening and followed by a normal day's operation; or votes were recorded prior to opening, continued throughout the day, and continued after polls were meant to have closed.
- Initially quarantined polling stations: IEC Decision 108 covers a batch of 137630 votes where there appear to be discrepancies between voters processed by a biometric voter verification device and individual biometric voter data for those stations. These were initially flagged by Dermalog based on discrepancies between a “processed voters” figure that is automatically generated and linked to a specific BVV device, and individual biometric voter signatures that are linked to a specific polling station on a digital SD card. Switching SD cards due to device malfunction (recording multiple station results on a single device, or using multiple devices to submit a single station) appears to be the primary cause of these discrepancies. In total, 4563 stations were reportedly affected, but no detailed list of which ones has been released; at 1466 polling stations there were 30,608 more voters in the device count than the SD voter count (so votes appeared to be “missing”), and at 3097 polling stations there were 168,238 fewer voters in the device count than the SD card voter count (so “extra” votes). The IEC broadly decided not to invalidate or audit votes under this criteria, concluding that most of the discrepancies were due to technical problems with the use of biometric devices and not evidence of wrongdoing. As noted above, only the latter category of stations appears to have been selected for sample audit on the basis of the ECC's special audit order. Beyond the list of stations sampled for audit on the basis of the ECC order, the IEC has not released any further information identifying all of the affected stations.
The IEC's publication of the list of invalidated or closed polling stations left ten polling stations not included in either the preliminary results data or on the list of confirmed invalidated or closed stations. (These stations are highlighted in the prelim_ps_missing_status file, or by ps_open status "UNKNOWN" in the polling station key report.) All of these stations were selected in the first round of audits, and seven were selected for further audit on the basis of the ECC order; four were added to the final results, as noted above.
Feedback, corrections, or suggestions for further expansion or collaboration are greatly appreciated. For questions, suggestions, or to contribute further, please leave an issue request here on Github or contact Colin Cookman by email or Twitter.
Although I am solely responsible for any errors in this project, I am grateful to Asma Ebadi and Lucy Stevenson-Yang for research and coding assistance; to the Independent Election Commission for releasing elections data in a public form that allows for open analysis; to the anonymous election observer sources who generously provided additional data sources; to the Afghanistan Elections Data, Map Sync, and OCHA Afghanistan / Humanitarian Data Exchange projects for their work on collecting other supplementary data; to the researchers and authors at the Afghan Analysts Network for their detailed reporting and analytical resources which have informed my understanding of Afghanistan's elections; to John Ray, Luke Sonnet, and the Stackoverflow and RStudio communities for lessons in the use of R for data collection, cleaning, and analysis; to Democracy International for providing the opportunity for me to gain firsthand experience as an observer to the 2010 and 2014 Afghan elections; and to Belquis Ahmadi, Staffan Darnolf, Scott Smith, Andrew Wilder, Scott Worden and many other colleagues at the U.S. Institute of Peace for their guidance, support, mentorship, and patience as I worked on developing this project.