A growing research literature has examined the role and impact of election monitoring by international organizations, exemplified by the activities of the Organization of American States (OAS), the Organization of Economic Cooperation and Security (OSCE), and the Carter Center. This body of work has focused on the impact of observers on deterring fraud in local polling stations, as well as the broader impact, and the development of multiple electoral observer missions from non-Western countries (Hyde 2011; Kelley 2012; Donno 2013).
By contrast, other than occasional case-studies, far less work has examined and compared the role of election monitoring by domestic watchdog organizations in civil society, although agencies in the international community, such as the National Democratic institute, have invested considerable resources in expanding the capacity of such groups.
In particular, NGOs are increasingly utilizing information and communication technologies such as smartphones and social media which enable a growing number of citizens to document the electoral process and to share their observations with their peers in near real-time. In a time of increasing mobile telephony and social media use, chances are that some citizens will witness intimidation, vote-buying, ballot box stuffing, miscounting of votes, or erroneous voter registration lists, and spread these observations via Twitter, Facebook, SMS, or one of a growing number of dedicated websites.
Such 'crowdsourced election monitoring' is hailed by enthusiasts as the way to enhance electoral integrity by unleashing the untapped potential of the ‘crowd’ to monitor wrongdoings and to hold electoral authorities as well as political parties and candidates accountable. Yet others caution against the dangers of hate speech and viral spread of misinformation, or simply expect more incremental change if any. At present, little systematic evidence is available to evaluate the merit of these rival claims.
The internet and mobile phones are reaching an ever increasing number of people, including in many low and middle income countries. Despite a persisting digital divide, 40% of the global population is now online and the number of mobile phone subscriptions exceeds the world population (ITU 2013). Mobile phones facilitate mobilization, information acquisition, and everyday political talk online. New communication technologies have rapidly shaped elections from online campaigning, to internet voting, voter registration by SMS, voting advice applications (VVA), campaign finance monitoring, to the tracking of candidates' popularity via online social networking sites.Taking these developments into account, the EIP research program on crowdsourced election monitoring has four primary goals:
The program aims to assess the trends of domestic monitoring with special attention to the technological 'affordances' (Earl and Kimport 2011) that enable many of these initiatives. Social networking sites, dedicated online mapping initiatives, or mobile phone apps extend the number of election observers potentially to the whole population. This can increase the amount of information as well as public participation. Domestic election monitoring efforts are situated along a continuum from organizationally-brokered collective action to crowd-enabled connective action (Bennett & Segerberg 2013).
The changed electoral information environment in the digital era can be captured by the concept of crowdsourcing. Derived originally from business contexts, it can be defined as "the act of taking a job traditionally performed by a designated agent […] and outsourcing it to an undefined, generally large group of people in the form of an open call" (Howe 2008:99). A few sparse contributions address this growing trend of "popular election monitoring" (Fung 2011), 'crowdsourced election monitoring' (Bailard & Livingston 2014) or "citizen election observation" (Tuccinardi & Balme 2013). The main advantage of the ‘wisdom of the crowd’ is said to lie in its diversity and its greater number of observations and thus greater coverage. Thus, the more observations are accumulated, the larger and more diverse the number of observers, and the more cross-checks are possible between individual observations, the larger should be the increase of the knowledge base available on electoral events.
Yet it must be noted that criticism about the inaccuracies or even dangers of crowdsourcing should be taken seriously. Lorenz et al. (2011:5) point out that “it is remarkable how little social influence is required to produce herding behavior and negative side effects for the mechanism underlying the wisdom of crowds”. This is the case if there is a strongly voiced opinion which can turn into majority opinion. In the context of elections this could happen in the form of deliberately false reports, or attempts by political parties to discredit their opponents.
In practice, there are at least three different types of crowdsourced monitoring in the context of elections. These distinction builds on Meier (2011) and Sambuli et al. (2013). They also show large overlap with Bennet's and Segerberg's typology (see above). The strategies differ mainly in the extent to which they engage with citizens. The first two types can also be called 'dedicated crowdsourcing', because they involve an active call for participation placed by an organization. This could be either an international or domestic election monitoring group, a media organization, or an election management body (EMB). The third type of crowdsourcing on the other hand can happen without any such dedicated call.
‘Bounded’ crowdsourcing: Reporting by trusted volunteers and staff
EMBs or election watchdog groups already have a wide network of staff and volunteers in place which can in effect be turned into a ‘bounded crowd’ of election reporters. Bounded, because the criteria for inclusion into the 'crowd' are clear and the EMB or monitoring group can train them in a clear reporting methodology. This type of crowdsourcing is a technologically enhanced version of what EMBs and election observers already do. A bounded crowd provides a ‘trust network’ (Lederach 2012), which is one way to increase the credibility of incoming reports on election risks. Since the ‘bounded crowd’ is trained in identifying electoral risks and adheres to a certain protocol for reporting, incoming information is more reliable than reports from the general public. Yet, a larger element of crowd scrutiny is intrinsic to such monitoring efforts, if they offer the observations back to the general public in the form of an online map.
A methodology that has proved particularly efficient in this regard is SMS messaging. It facilitates the quick delivery of reports from trained field observers, who might be spread over hundreds of polling station. It can be implemented by relying on a mass-SMS messaging service such as FrontlineSMS, Clickatell, or the Carter Center’s open source Election Monitoring (ELMO) tool. These tools facilitate both the handling of incoming messages as well as automated responses. Ideally, reports should be received in a particular SMS syntax, meaning a pre-determined reporting format that encodes different incidents into an (alpha)numerical code. Not only can observers quickly send such standardized SMS at regular intervals to keep HQ abreast of developments. Coded text messages can also be automatically read into a database, instead of being received and processed manually one by one. If the syntax includes a geographical reference, this information can easily be translated into a map. Reduced costs and timeliness of reporting are seen as great advantages of SMS use (Schuler 2008; Bardall 2010).
‘Open’ crowdsourcing: Reporting by citizens
Going one step further than SMS reporting by a ‘bounded’ crowd is the inclusion of the whole population in a large open crowdsourcing effort. One can understand crowdsourced election monitoring as a system in which "any individual can register an observation about an election [which is] is pooled with other individuals' observations to create a public depiction of the reality of the election that is offered back to the public and to election officials in real-time on election day" (Fung 2011: 194-195). Chances are that some citizens will witness incidents of intimidation or fraud and can provide valuable information on them. In practice, ‘open crowdsourcing’ means that an open call is placed to the entire populace to report such instances via email, twitter, social media, phone call, SMS or a combination of channels. The information is then verified by different methods and usually displayed publicly in an online platform, often in the form of a map. Different software solutions exist, but many projects make use of the open source Ushahidi platform combined with text messaging services.
Most of the time, such initiatives are implemented by election watch NGOs, who hope to increase both their information base and their popular appeal by involving the citizenry-at-large. EMBs remain more adversarial (an exception is the Reclaim Naija platform - see below). Yet, the method presents some advantages for EMBs as well. First, communications with EMB staff cannot cover all locations (especially remote polling places), and poll workers are only present from shortly before to shortly after the election. Open crowdsourcing on the other hand enables citizens to report incidents of intimidation or vote buying even in remote locations and weeks before or after the elections. This information becomes thus accessible to the EMB for its security planning. Second, giving voice to citizens themselves might increase the sense of popular ownership of the electoral process. Transparency and ‘open government’ are rapidly becoming standard practices around the world in other governance sectors. At the same time, EMBs often have to face the brunt of citizen anger in the aftermath of problematic elections. A proactive approach of engaging the citizenry in gathering information on problems might signal the EMB’s commitment to electoral integrity and increase public trust.
Two main challenges of open crowdsourcing remain the question of credibility of incoming reports, and how to turn these into actionable information.
‘Passive’ crowdsourcing: Listening to social media
While the previous two approaches rely on actively sourcing reports from either a bounded or an open crowd, so-called ‘passive’ crowdsourcing is the gathering of information “produced as a result of the existing behavior of a crowd of users on existing platforms, such as Twitter” (iHub Research 2013:11). It is passive in the sense that there is no active call placed to the population or trained observers. Instead, it makes use of the fact that elections are highly politicized events, and citizens are likely to talk about them in some detail on social media. An election observer group or and EMB can listen to these conversations as they occur spontaneously on Twitter, Facebook, Instagram, Tumblr and other social networking sites.
Election monitoring groups and EMBs face difficult challenges in the implementation of passive crowdsourcing. Unlike the previous two approaches which ask a precise question of observers or the whole population, passive crowdsourcing is like eavesdropping in on a conversation without knowing exactly what to listen for. So first, relevant search terms and hashtags need to be identified, which can then be implemented in a systematic and continuous search on social media. This aids in building a ‘corpus’ of relevant posts on social networking sites which can then be analyzed either manually or in an automated way. Examples of Twitter hashtags can be intuitive such as #zimelections2013 or #ghanavotes. On the other hand, trending topics do not necessarily conform to previously known or set search terms. The nature of online talk is that it is in constant flux, and thus the corpus needs to be updated continuously. This requires either a lot of volunteers, or machine-learning techniques.
In addition, a strategy needs to be implemented for the verification and analysis of the mountains of monitored information. Not only will the signal-to-noise ratio be unfavorable due to the rather unstructured nature of posts. There is also the huge challenge of separating rumors and deliberate slander from fact.
The program is interested in the effects of monitoring on electoral integrity. Recognizing that large-scale structural factors of economic development, state capacity, geography, international engagement and others strongly influence overall levels of electoral integrity within countries (Norris 2015), and that regime-level logic is likely to drive the overall extent of electoral manipulation (Schedler 2013, Simpser 2013), it is likely that effects of monitoring are only marginal - yet they should be measurable. The discussion of effects is embedded in a larger theoretical framework of participation, transparency, and societal accountability (Smulovitz & Peruzzotti 2000).
Elections can be seen as an instance of public good provision, yet they are distinct from other public policies in several important ways that have implications for an accountability framework. First, prospective or retrospective electoral accountability is in itself the underlying assumption in most principal-agent models (Ashworth 2012). Yet, this very mechanism is threatened by electoral malpractices or electoral manipulations. Second, unlike some well-delineated public policies, elections are produced by a 'governance network' (Sørensen & Torfing 2007) consisting of the election management body (EMB), political parties and candidates, security agencies, the media, and the voters themselves. No single actor is solely responsible for their planning and implementation, and the de facto influence of EMBs on the conduct of elections might be much more limited than commonly assumed. Third, clientelism - endemic in many of the same countries that are affected by electoral malpractices - undermines the accountability function and turns it into a function of patron-client responsiveness. From the perspective of principal-agent theory there might be several different principals who try to influence agents' (bureaucrats' and politicians') actions. Thus, monitoring might have differential effects.
Following the literature on transparency and accountability interventions, a causal chain model of possible effects is proposed:
a) Information on electoral malpractices is provided by domestic monitors; b) public perceptions of the integrity of the election are altered if this Information is accessible, salient and credible to voters; c) voters may take action to put pressure on cheating parties and on election authorities (protest, petition, increased turnout) under certain conditions; if then d) EMBs and political parties are responsive to such public pressures; then e) electoral malpractices are deterred (first-order effect) or electoral reform is initiated (second-order effect).
Fig.1: Causal chain model of effects of crowdsourced election monitoring
At each step of this causal chain, many conditioning factors have to hold in order for change to occur. Alternative explanations need to be considered and tested. The default assumption is that no change takes place whatsoever. Considering all effects along this hypothesized causal chain is a long-term research agenda.
In the first phase of the research program, effects on public perceptions of electoral integrity - step b) in the above model - will be assessed at the individual and the aggregate level. In essence, this means testing the assumption of the causal chain model that citizens are interested in information about electoral malpractice, and that such information changes their perceptions of the integrity of the election if it is credible and accessible. But this assumption needs to be unpacked and tested. Alternative explanations would for example include the overwhelming dominance of elite cues (e.g. political parties or ethnic groups ), an immutable public perception (due to disinterest for example), or a lack of cognitive capacities to deal with informational abundance, just to name a few.
Some of the factors to consider are listed below:
First, the persisting digital divide influences who has access to information provided by domestic election monitors. Of particular importance for more crowd-enabled efforts are internet and mobile phone penetration rates.
Second, even if different communication channels are accessible, individual-level characteristics of audiences (such as interest in politics, education, socio-economic status) are likely to influence the extent to which they use media to form a view on the conduct of the election.
Third, election-level characteristics should shape public perceptions of electoral integrity. In particular, the actual integrity of the election, as well as the closeness of the race should have strong effects.
Fourth, it is possible that information from domestic watchdog groups lacks the legitimacy and credibility needed to affect public perceptions. On the one hand, authority of the sender traditionally played a huge role in credibility assessment. But "bottom-up assessments of information quality can be constructed rather easily through collective or community efforts enabled by technology, which in some cases allow information consumers to bypass traditional authorities altogether" (Metzger, Flanagin & Medders 2010:434). Hence, it is an open question whether organizationally-enabled monitoring efforts are more credible because they rely on expert knowledge and authority, or whether crowd-enabled efforts are more credible because they generate information in a participatory and expressive way.
Fifth, contradictory or two-sided messages about the quality of an election are likely to produce differential effects, depending on the intensity of the messages and prior opinions (Zaller 1992). In particular, partisans of the winning party might not be interested in any information on fraud. This might be exacerbated through selective exposure to information from channels that are co-partisans. In this regard, it is of great importance whether the monitoring initiative is perceived as being partisan or neutral itself.
Each of these factors is taken into account in a step-wise estimation of the effects of domestic monitoring on public perceptions of electoral integrity. Of special interest are characteristics of the election monitoring initiatives, namely their partisanship and their level of 'connectiveness' .
Existing and bespoke data sources are used to test the effects of crowdsourced election monitoring on electoral integrity. Important existing data sources are the Perceptions of Electoral Integrity (PEI) dataset, the World Values Survey (WVS), and the National Elections Across Democracy and Autocracy (NELDA) dataset.
In addition, a novel dataset on domestic election monitoring is developed - the Crowdsourced election monitoring dataset (CEMD). Temporal coverage at present is from 2010-2014. Relevant monitoring projects are identified by triangulating the database of the Global Network of Domestic Election Monitors (GNDEM) with a keyword search on a number of online databases that systematically collect information on citizen media around the world. Variables in the CEMD include location, type (bounded, open, passive), data-validation strategy, linkages with state and civil society, number of reports on website (if any), resonance in social media platforms and others.
The CEMD is used to derive two variables of interest for later usage. First, the initiatives are located on a continuum from 'organizationally brokered networks' over 'organizationally-enabled networks', to 'crowd-enabled networks' (Bennett & Segerberg 2013). This classification overlaps to a large extent with the distinction between 'bounded', 'open' and 'passive' crowdsourcing.
Furthermore, social network analysis is used to code the degree of partisanship of the domestic monitoring initiatives. Bipartite page-user interaction networks are extracted from the facebook pages of the domestic election monitoring groups. These are overlaid with the interaction networks of the facebook pages of the ruling and the main opposition parties in each country. The overlap of the combined networks gives evidence as to whether the online publics are partisans of opposition parties.
The progress of the CEMD database is continuously updated and mapped. This map is a work in progress. It depicts a selection of dedicated open and bounded crowdsourcing websites that collect information on electoral manipulations or election violence.
IF YOU KNOW OF OTHER CROWDSOURCED ELECTION MONITORING PROJECTS, PLEASE CONTACT max.groemping[at]sydney.edu.au
In future phases of the research program, the other downstream effects theorized in the causal chain model will be discussed. In particular, effects on political behaviour, on the responsiveness of political parties and electoral authorities to public pressures, and net effects on electoral malpractices. This will necessitate the development of further data. In particular, a survey of the online publics of crowdsourced monitoring initiatives will be necessary to assess individual-level effects on the behaviour of the audience. Similarly, survey data on election officials will be needed to look into the blackbox of EMB responsiveness. Finally, randomized controlled trials (RCT) are to be used to estimate the average treatment effects of either physical presence of election observers or the presence of online reporting channels on election fraud or administrative malpractices.
An explicit aim of the EIP program on crowdsourced election monitoring is to identify best practices and develop actionable recommendations that can enhance electoral management and electoral assistance. For this purpose, outreach and networking with election monitoring groups and election management bodies are priority activities. Examples are practitioner events such as the Secure and Fair Elections (SAFE) workshop organized by EIP, UNDP, and International IDEA, with the participation of the election commissions of Afghanistan and Nepal. Furthermore, scholarly events such as the 2013 workshop on 'Electoral Integrity in Southeast Asia' (attended by ANFREL, COMFREL, The Asia Foundation, Bersih, and LENTE), or the 2014 workshop on 'Electoral Integrity in Asia-Pacific' (attended by the Australian Electoral Commission, UNDP, IFES, and The Carter Center) facilitate the exchange of knowledge between academia and practitioners. These events help grounding the scholarly research in the real-world experience of organizations that use crowdsourcing in their programming. Simultaneously, the program's empirical research findings can be used to inform practitioners' operations on the ground.
Nigeria (2011-2014)During the Nigerian 2011 elections, for example, the open crowdsourcing website Reclaim Naija was used by citizens from all over the country to report incidents during the voter registration exercise. The reports were collated in real time and fed back to the Independent National Electoral Commission (INEC). This assisted INEC in troubleshooting in many locations across the country. The Commission deployed personnel and distributed election material based on reports on the website. The information was also useful to the media in monitoring and publishing stories on the voter registration exercise, thereby helping to amplify the voice of the people. In addition, the reclaimnaija.net website serves as a one-stop online resource for information on the 2011 elections. It featured all the polling units, senatorial districts and wards, the Nigerian Constitution, information on candidates, the 2010 electoral act, the election timetable, electoral guidelines, certified voters’ registration figures, political parties as well as civic and voter education modules.
The website has been in continuous use since then - for example during state elections in 2014.
An instance of open crowdsourcing on a dedicated website is the experience of the independent Russian NGO, the Golos Association, which launched the Map of Electoral Violations following the contested Duma elections in 2011. Since then the organization has used their website Karta Narushenyi to monitor a series of Russian elections for local, mayoral, Duma and presidential contests. Users can post comments, documents and videos on the website, through online messages, SMS and telephone hotlines.
The organization's analysis of the March 4th 2012 Presidential elections, for example, noted numerous cases where there were problems for electoral observers and media reporters, irregularities in absentee voting, and violations of vote-counts, especially in Moscow and St Petersburg.
Based on this source and other evidence, the organization concluded that the Russian elections were flawed by: “an insufficient level of competition, government interference with the electoral process, and some degree of coercion to vote. These elections cannot be classified as free and fair under the definitions provided in the Russian constitution and international electoral standards.”
The platform has been used for subsequent elections in the Russian Federation.
A different approach was taken by the initiative Senevote, developed by the Senegalese election watch coalition COSCE in partnership with OneWorld and the Open Society Initiative for West Africa (OSIWA). Senevote is an example of bounded crowdsourcing, in which COSCE's trained observers constituted the 'crowd' from which reports were coming in. Mobile technology and online mapping was in this case an extension of the established methodology of domestic election observers. SMS reporting streamlined and sped up the reporting process and led to the incredible amount of about 74000 individual observation at the polling station level.
In addition to bounded crowdsourcing, the initiative also pioneered the situation room set-up, that has since been implemented in other African countries, and - thanks to a GNDEM exchange program - as far away as Cambodia. The situation room is meant to provide a one-stop shop for election-related information, bringing together all electoral stakeholders. In the ideal-typical set-up, civil society watchdogs, election authorities, security services and the media, would be able to share information on and around election day instantly. The presence of decision-makers would allow for a rapid response to arising integrity threats.
Yet another innovative approach was taken by the Video The Vote project during the 2012 presidential elections in the US. A collaboration between a diverse range of civil society and citizen-action organization dating back to 2006, the project placed an open crowdsourced call to the citizenry to document election irregularities via video. It included information on election laws and guidelines on what to look for - such as long voter lines, intimidation, (in)accessibility of polling places, or voters being turned away from the polls.
For local elections in Turkey, a group of citizen journalists initiated the project 140Journos, named after the character limit for Twitter posts. The group employed passive crowdsourcing by monitoring Twitter, facebook and other online social networking sites. On election day, the group operated an 'operations room' and aggregated social media reports into infographics and maps. In addition to monitoring irregularities, the group also attempted a parallel vote tabulation (PVT) via 'photo quick count' by soliciting pictures of vote count results from the polling station. This methodology had previously been used as a form of monitoring in Afghanistan and Uganda.
The project was continued for the presidential election in August 2014.
The example of the derailed Thai general election of February 2014 showcases how social media monitoring can provide valuable and actionable information through passive crowdsourcing. On advance voting as well as on election day, the opposition movement hindered voters in numerous polling places from entering and casting their vote – which in the end led to the annulment of the election. A wave of images and videos went viral on Facebook and Twitter depicting voters being harassed, attacked, or disappointed to tears when they could not vote. Others showed voters defying the protesters by climbing fences and attempting to vote anyway (Grömping 2014).
Such images and videos constitute valuable information and – if verified – could be used by EMBs to dispatch staff, alert the authorities or take action in some other way. This can prevent further intimidation or simply enhance voters’ experience of the electoral process. Unfortunately, in the Thai case, the information remained unused and the Election Commission of Thailand was criticized for its inaction in the aftermath.
Collection of social media memes in the 2014 Thai election (Source: Facebook)
Indonesia(2014)In the Indonesia presidential election of 9 July 2014, different polling agencies announced contradictory election results. Consequently, both candidates claimed victory, and a battle over public opinion and whose count to trust ensued. The Indonesian EMB had previously committed to placing scanned images of the official vote tallies from every single polling station around the country online. In response to the perceived confusion over different result polls, several open crowdsourcing initiatives were started by technologists and a large number of online volunteers to conduct parallel vote tabulations. They scraped all scanned counts off the EMB website, and asked internet users to translate them into machine-readable format. Two examples are Kawal Pemilu and Kawal Suara This count confirmed one of the candidates and dispelled the polling results of the other one. It was an important factor in swaying public opinion and conferring legitimacy to the winning candidate.
Academic publications and working papers
Other related online resources
List of International Monitoring Organisations
For further information, contact the program manager:
Electoral Integrity Project
The University of Sydney
Department of Government and International Relations
Merewether 260 (H04) | The University of Sydney | NSW | 2006
phone: +61 (0)2-935-15085
skype: mgroem / twitter: @MaxGroemping
 See for instance the Internet Voting Project for more information.
 See for example the voter registration process in South Africa.
 E.g. Art61 in Poland, Thailand Political Database in Thailand, Abgeordnetenwatch in Germany, www.votebd.org, Vota Inteligente in Chile, uElect in Pakistan, Ayo Vote in Indonesia, Chesno in the Ukraine, or Cine ca a promis in Romania.
 Such as the Argentinian initiative Dinero y Politica, Lobby Control in Germany, or the Hungarian Képmutatás.
 'Crowdsourced' because all domestic monitoring groups rely on a network of volunteers. They thus rely on the 'crowd' and connect with it via technology to varying degrees.