Abstract
Low state capacity makes it difficult for governments in developing countries to deliver resources to the poor. In this chapter, we highlight the role that biometric authentication can play in delivering payments and subsidized food to the poor. We describe the implementation and evaluation of two different biometric authentication systems in Andhra Pradesh (“AP Smartcards”) and Jharkhand (“Aadhaar”), India. Results from two large-scale RCTs (Muralidharan et al., 2016 and Muralidharan et al., 2020b) showed that more accurate biometric ID systems, coupled with payments and policy reforms, reduced leakages in welfare schemes in both Andhra Pradesh and Jharkhand. However, there were varying results on beneficiary welfare. In Jharkhand, reduced fiscal leakage came at the expense of excluding genuine beneficiaries who were unable to meet new standards for identification. Exclusion of beneficiaries was low in Andhra Pradesh, where the government was more focused on improving beneficiary experience with welfare programs. The studies discussed in this chapter highlight how differences in policy priorities and the details of solution design influence the extent to which beneficiaries benefit from biometric authentication and accompanying reforms.
The authors were affiliated with the Payments and Governance Research Program in the capacity of Research Associate and Project Director, respectively. We thank Karthik Muralidharan, Paul Niehaus, Sandip Sukhtankar, and the Payments and Governance Research Program for their support. In addition, we would like to thank Anustubh Agnihotri, Temina Madon, Karthik Muralidharan, Arman Rezaee, and Sneha for their comments and suggestions.
You have full access to this open access chapter, Download chapter PDF
Similar content being viewed by others
Keywords
- Aadhaar
- Biometric authentication
- Corruption
- Digital identification
- Inclusion and exclusion errors
- India
- Leakage
- Public distribution system
- Public service delivery
- State capacity
1 Development Challenge
Throughout the world, government welfare programs face two intransigent challenges: (1) how to deliver public assistance to the neediest, the “targeting” problem, and (2) how to minimize the “leakage” of resources from public coffers (World Bank, 2003). Targeting assistance to those in need is particularly difficult in developing countries because government agencies often lack reliable data on household income and government systems for welfare monitoring are weak (Niehaus et al., 2013). There is also widespread corruption in the delivery of services, a problem documented by many researchers over several decades (Bardhan, 1997; Olken & Pande, 2012). Corruption in public services can come in many forms, from skimming off beneficiary payments to creating “ghost” beneficiaries and extracting bribes from eligible beneficiaries. In 2006, Olken found that 18% of subsidized rice distributed to poor families in Indonesia was unaccounted for, when comparing household reports with administrative data. Even more stark, a study in Uganda estimated that on average just 13% of reported government expenditure on primary schools was actually delivered to schools (Reinikka & Svensson, 2001).
The diversion of public resources away from intended beneficiaries is a development challenge on multiple fronts. First, the social and economic value of public welfare programs is drained when those dependent on assistance are prevented from accessing their entitlements (Ferraz et al., 2012; Niehaus & Sukhtankar, 2013). Second, leakage of public funds substantially reduces the efficiency of governments. This is particularly detrimental for developing countries that have limited fiscal capacity, and it undermines citizens’ confidence in the public sector (Muralidharan et al., 2017; World Bank, 2003).
Our case study focuses on India, where it is estimated that a significant proportion of spending on key forms of public assistance does not reach its intended beneficiaries (see Table 20.1, from the 2014 to 2015 Indian Economic Survey and Dreze & Khera, 2015). It is estimated that these leakages cost the Indian government an estimated Rs. 28,534 crores annually – translating to roughly 4.7 billion dollars, or 0.2% of India’s GDP.
Digital identification (ID) technology offers a promising solution to the challenges of public welfare distribution. In this chapter, we examine the application of two different biometric identification systems in India: a “Smartcards” pilot in the state of Andhra Pradesh and the use of Aadhaar, the national digital ID, in the state of Jharkhand. The applications combined advances in biometric authentication technology with a suite of policy reforms, public-private partnerships, and service delivery innovations. Taken together, these innovations represent a profound investment in India’s state capacity.
To grasp the novelty and significance of India’s Aadhaar, it is helpful to first understand how most governments operate welfare programs today. Welfare services are typically “targeted” to those most in need, with eligible beneficiaries identified through proxy means testing (PMT). These tests establish eligibility for assistance based on readily observable household characteristics that closely proxy income – such as household size, land holdings, quality of housing, and consumer durables. PMT models or algorithms are designed and validated using data from representative surveys (which include more detailed measures of household income, assets, and consumption). A household’s score is usually assigned by government workers, who directly observe the outcomes that determine eligibility.
However, targeting is prone to two types of errors: (1) inclusion errors, giving benefits to households who are not eligible for the program thereby increasing leakage and fiscal burden on the government, and (2) exclusion errors, denying benefits to eligible households.
In India, household scores are assigned at the state level, and those falling below the government’s poverty line are issued a below poverty line (BPL) ration card (Alkire & Seth, 2013; Niehaus et al., 2013). To access public benefits, registered individuals or households have traditionally presented their BPL card alongside documents that authenticate their identity. However, many people lack this documentation, particularly households and individuals that are socioeconomically marginalized or disadvantaged. It is estimated that over a billion people in the world lack legal identities (World Bank, 2019a).
Assuming eligible citizens or households are correctly identified, the next challenge is reaching and authenticating the targeted beneficiary. Recent advances in biometric technology present a solution to the challenge of rapidly, accurately, and uniquely identifying welfare beneficiaries (Gelb & Metz, 2018). Biometric technologies capture and digitize the physical traits of individuals including fingerprints, iris scans, and speech. As these traits are unique and attached to a person, they cannot be lost or forgotten and are difficult to replicate (unlike paper identification cards; see Goldberg et al., 2010). In addition, ID programs that use biometric technology are more suitable for populations with low literacy and numeracy, particularly compared to alternatives like passwords and pin numbers (Muralidharan et al., 2016).
In the past two decades, the number of national ID programs that use biometric technology has grown rapidly (Gelb & Metz, 2018). India’s ambitious Aadhaar initiative was one of the first, providing unique biometric IDs to over a billion residents in under a decade. Similar large-scale initiatives are currently underway in Kenya (Huduma Namba), Indonesia (Kartu Tanda Penduduk Elektronik), Nigeria (National Identification Number), and the Philippines (PhilSys).
In addition to authenticating citizens’ identities, biometric technology can also be linked to the banking ecosystem, enabling governments to use existing payment systems to deliver welfare transfers. Money can be deposited directly into the bank accounts of beneficiaries, which creates more credible records for government accountability and transparency compared to cash-based systems. Even better are digital payments, which can be continuously monitored and audited to expose fraud.
With the proliferation of mobile phones, biometric IDs and bank accounts are also being interlinked with beneficiaries’ mobile numbers (i.e., SIM cards), making it possible for banks to remotely authenticate an individual’s identity, complete digital payments, and provide customer support without brick-and-mortar infrastructure. Digital payments can also be delivered to the doorstep of the beneficiary, by deploying mobile banking correspondents equipped with point-of-service (PoS) devices to authenticate users and then disburse cash on the spot.
While there are many potential benefits of digital ID technology, these systems also raise important concerns about citizens’ privacy and the exclusion of eligible beneficiaries. Many developing countries lack the legal safeguards, legal capacity, and civil society organizations needed to protect individuals from privacy violations and other abuses by the state. For example, there are concerns that biometric IDs can be used by governments to profile citizens and carry out mass surveillance activities, under the guise of national security (Dreze, 2016). There are also fears that governments can monetize interlinked databases, by selling data to private entities (Bhatia, 2018) without citizens’ permission.Footnote 1 In some countries, biometric ID enrollment efforts have failed to reach the most marginalized households, and it is unclear whether this is politically motivated or a failure of the bureaucracy. Those lacking official identification are made more vulnerable when biometric IDs are made mandatory for accessing welfare schemes and other government services like education and health care (Dahir, 2020).
The Aadhaar program in India is the largest application of biometric technology in a national ID program; and given its scale, it is a unique opportunity to study how the design of a new technology influences economic development. The effort has exposed a number of intertwined technological, bureaucratic, social, and ethical challenges. Technical problems like hardware failures and lack of Internet connectivity have led to the exclusion of genuine beneficiaries and the starvation deaths of a few. Government collection of biometric data en masse has raised important privacy and security concerns among members of India’s civil society (Dreze et al., 2017; Khera, 2019a, b). The opportunities for learning – both what works and what has failed – are unparalleled (Box 20.1).
Evolution of Biometric IDs in India
Calls for a biometric identity card in India can be traced back to national security concerns related to illegal immigration across India’s borders (Government of India, 2001). In 2003, the government launched a biometric Multi-Purpose National Identity Card (MNIC) for all citizens. The MNIC project aimed to provide a “credible identification system” for speedy delivery of public and private services and which could also serve as “deterrent for illegal migration” (Census of India, 2003).
With these aims in mind, a pilot of the MNIC was initiated in 2006 in 13 districts. The pilot ran for a period of 3 years but failed to establish citizenship for more than half of the residents of participating districts. Rural residents, with a weak document base, found it particularly difficult to prove their citizenship (Singh, 2020).
In parallel, the Indian government was developing a biometric identification project focused on identifying families whose incomes were below the poverty line (BPL), with the aim of improving targeting of public assistance to poor households. Challenges in establishing citizenship for these households – and the weak results of the MNIC pilot – led the government to shift its investment in biometrics away from citizenship and toward welfare targeting (Misra, 2019).
In the following years, different states in India introduced their own biometric ID programs to improve last-mile delivery of welfare schemes. Following the success of some state-level biometric ID programs including the AP Smartcards project, the union government created the Unique Identification Authority of India (UIDAI) in 2009, charged with implementing a nationwide biometric identity project. This came to be known as Aadhaar. The first Aadhaar card was issued in September 2010; and as of 2020, over one billion Aadhaar cards have been issued to Indian residents – making it the world’s largest digital identity program (Abraham et al., 2018).
Although the need for a biometric ID card initially emerged out of national security concerns, Aadhaar is now viewed by the government as a tool to reduce leakages in social programs and ensure better targeting of welfare schemes (Government of India, 2015).
1.1 Overview of the Case Study
In this case study, we will examine two different biometric identification systems in India: a “Smartcards” pilot in the state of Andhra Pradesh (AP) and the full-scale launch of the Aadhaar system in the state of Jharkhand. In each case, the implementation of the new technology was achieved through partnerships between state governments and private entities such as banks. Here we will describe the policy innovations and digital technologies used to form a new digital welfare state across India. We will also outline results from two large-scale randomized controlled trials (RCTs) conducted by the Payments and Governance Research ProgramFootnote 2 evaluating the social and economic impacts of these systems (Muralidharan et al., 2016, 2020b). The RCTs were designed to credibly estimate the impact of programs implemented by governments at scale. As a result, these studies also document the complex administrative, logistical, and social factors that can influence the implementation of large-scale technology initiatives.
The chapter is organized as follows: Section 2 provides contextual details of welfare schemes in India, including earlier methods of targeting and delivering benefits. Section 3 describes how biometric identification systems facilitate reforms in beneficiary authentication and service delivery. Section 4 outlines the main challenges in implementing biometric ID programs in India. Section 5 presents results from the large-scale evaluations in two states, and conclusions are provided in Sect. 6.
2 Implementation Context
Despite rapid economic growth since liberalizing the economy in the early 1990s, India’s development indicators remain poor. As of 2020, India currently languishes at 131st (out of 189 countries) in the United Nations’ Human Development Index (UNDP, 2020). The 2020 Global Hunger Index placed India at 94th out of 107 countries, far behind South Asian peers with slower growth rates. The country also continues to have one of the highest infant mortality rates in South Asia (second only to Pakistan) (UNICEF et al., 2020).
It is argued that India’s sub-par development indicators are partly a result of the government’s weak implementation of welfare schemes (Pritchett, 2009). Last-mile delivery of many government programs has been undermined by low state capacity (Muralidharan et al., 2016; Sukhtankar & Vaishnav, 2015). In this section, we motivate the introduction of biometric IDs in India. We describe common modes of failure in the nation’s service delivery and how these failures undermine India’s investment in poverty reduction and economic development.
2.1 Public Distribution System
The Government of India’s Public Distribution System (PDS), established at independence as a way to regulate food prices and prevent famines, provides subsidized food and fuel to over 800 million beneficiaries in India through a network of fair price shops (FPS). The Indian government spends approximately 1% of its GDP and 12% of its social services expenditure on the PDS (Government of India, 2020).
Beneficiaries of the PDS are assigned a ration card which lists household members and displays a photograph of the household head. They are then assigned to a geographically close FPS. They visit the FPS in person (ration cards in hand) to purchase rations each month at a highly subsidized price.
Prior to the introduction of biometric authentication, shopkeepers at the FPS (known as FPS dealers) were expected to record transactions on both ration cards and their own paper ledgers, neither of which were regularly audited by the government. The informal nature of authentication, alongside the dual price system created by public subsidies, created opportunities for leakage. This resulted in the diversion of PDS commodities to the open market, both directly from government warehouses and from trucking networks established to deliver goods to FPS dealers. In addition, beneficiaries have reported the adulteration of goods, overcharging for rations, and reduction in their entitlements at FPS (Muralidharan et al., 2020b).
2.2 National Rural Employment Guarantee Scheme
The National Rural Employment Guarantee Scheme (NREGS) was implemented across India after the enactment of the National Rural Employment Guarantee Act in 2005. The Act mandated that state governments set up employment programs guaranteeing 100 days of paid employment per year to any rural household. The Indian government spends approximately 0.3% of its GDP and 4% of its social services expenditure on the NREGS (Government of India, 2020).
Work provided through the NREGS can vary across villages but typically involves minor irrigation work or improvement of marginal lands. There is no eligibility criterion for the scheme, and those in need of wage labor can access the scheme (although the poorest households are more likely to self-select into the program). Households that participate in the scheme receive a job card which lists household members and has empty spaces for recording employment and payment. The job cards are issued by the local village or sub-district government offices. Workers with job cards can apply for work at will, and officials are legally obligated to provide either work or unemployment benefits (although, in practice, the latter are rarely provided; see Muralidharan et al., 2016).
Worksites are managed by officials called Field Assistants, who record worker attendance and output on “muster rolls.” These are then sent up to the sub-district office for digitization, after which the work records are sent up to the state level. This triggers the release of funds to pay workers. The state government transfers money to district offices, which then pass the funds to sub-district offices, which in turn transfer it to the beneficiary’s post office savings accounts at the village level (see Fig. 20.1).
Payment mechanism of the NREGS prior to the roll-out of biometric authentication
Note: The black arrows denote the flow of information on beneficiaries of the NREGS. The arrows in blue denote the flow of funds to the beneficiary. Source: Muralidharan et al. (2016)
Prior to the roll-out of biometric authentication, workers would withdraw funds by traveling to their village’s post office, where they would establish their identity using job cards and bank passbooks. In practice, it was common for workers, especially those who were illiterate, to give their documents to the Field Assistant who would then control and operate their bank accounts. This would entail taking sets of bank passbooks to the post office, withdrawing cash in bulk, and returning to distribute it in villages.
The payment mechanism was susceptible to leakage of two forms – over-reporting of hours worked and underpayment – with the former being more prevalent (Niehaus & Sukhtankar, 2013).Footnote 3 Two extreme forms of over-reporting are “ghost” workers who do not exist, but against whose names work is reported by Field Assistants, and “quasi-ghost” workers who exist, but who have not received any work or payments, even though their work is reported and payments are made. In both cases, the payments are typically siphoned off by government officials (Muralidharan et al., 2016). Estimates of leakages in the NREGS have varied across districts in India, ranging from 5 to 40% (Mookherjee, 2014). Additionally, payments made to beneficiaries through the NREGS have also been slow and unreliable. In some extreme cases, delayed payments have led to worker suicides (Pai, 2013).
2.3 Social Security Pensions
The Social Security Pensions (SSP) scheme, part of the Indian government’s National Social Assistance Programme, offers monthly payments to vulnerable populations. Eligible households must live below the poverty line (BPL), meet an additional vulnerability criterion,Footnote 4 and not be covered by any other pension scheme. Lists of eligible households are prepared by local village assemblies and sanctioned by the administration at the sub-district level. Pensions pay Rs. 200 per month except for disability pensions, which pay Rs. 500. A fund flow similar to that illustrated in Fig. 20.1 existed for the SSP (Muralidharan et al., 2016).
As the scheme mostly has a fixed list of beneficiaries who received a fixed payment, at a fixed time every month, for every month of the year, the SSP has been less prone to leakages than the NREGS and the PDS (ibid.). Leakages in the SSP have mainly taken the form of postmen disbursing less cash than the beneficiary’s entitlement, the existence of false beneficiaries (including deceased persons), demand for bribes, or theft. A study in the state of Karnataka found leakages in the SSP to be relatively lower, at about 17% (Dutta, 2008; Dutta etal., 2010).
3 Innovation
Biometric authentication technology has the potential to transform the delivery of public services in resource-poor settings, where state capacity is limited. It promises to reduce leakages resulting from ghost beneficiaries and improve the quality of last-mile service delivery. In the case of India, the technology’s adoption took place over a decade, beginning with initial state-level pilots and eventually scaling up across the country. This case study covers the pilot phase, in the form of an offline smartcard biometric system in the state of Andhra Pradesh (referred to as “AP” or “AP Smartcards”) between 2010 and 2012, as well as the central government’s fully scaled program, Aadhaar, as implemented in the state of Jharkhand in 2016–2017.
Both state governments introduced biometric ID systems alongside other technology advances and policy reforms. For example, they linked biometric IDs to bank accounts and to the existing databases of large welfare schemes, complementing the central government’s push to expand financial inclusion and improve service delivery (Government of India, 2015). However, the two states differed in the application of technology: AP used biometric IDs for rural employment (NREGS) and social security pensions (SSP), while Jharkhand used biometric IDs for food and fuel subsidies (PDS). The resulting solutions are described here. We first explain how the solutions modified government identification protocols and then describe how payment systems and service delivery were reformed.
3.1 Reforms in Identification
As discussed earlier, failures in paper-based government identity systems have allowed for several forms of leakage from public programs – from the creation of “ghost” welfare beneficiaries and the exclusion of intended beneficiaries to the extraction of bribes. Biometric authentication enables more accurate, auditable, and secure identification and targeting of beneficiaries.
To establish a biometric ID system, individuals must first be registered or enrolled. Their physical traits are recorded using devices that contain cameras, scanners, or other hardware. A backend software system then extracts, encodes, and stores this information in a database. Any request for authentication of a registered user compares an “input” provided in person (like a fingerprint) against the database of encoded biometric data (see Fig. 20.2). As the entire process is automated, authentication takes only a few seconds in most cases, including for Internet-connected systems (Xavier et al., 2012).
An important design choice is around the storage location for an individual’s biometric data. It can be stored centrally on a server or stored locally on a smartcard.Footnote 5 If biometric data is stored on the card, authentication can occur offline. A card reader is used to confirm that the data stored on chip matches the input presented in person. However, this approach has a number of limitations. First, each smartcard requires a chip for storing data, making large-scale programs quite expensive (World Bank, 2019b). Second, without a central database that assigns a unique ID to each individual, the biometric data stored on a smartcard is prone to duplication, which can be difficult to discover and remediate. Therefore, stand-alone biometric smartcards are prone to some of the same issues that affect traditional paper-based identification processes (Banerjee & Sharma, 2018). Lastly, smartcards place limits on usability, as the physical smartcard must be presented for access to services. If a beneficiary loses or damages their smartcard, there is no backup: they are now unable to authenticate themselves.
At the time the AP government initiated its biometric ID project in 2007, there was no universal ID database, and high-speed mobile networks were not universally available. As a result, the government chose to issue smartcards with local storage of data. This meant using biometric data to authenticate welfare beneficiaries only at the point of benefit disbursement (Mukhopadhyay et al., 2013).
In the case of Jharkhand, the state was able to use a newly created central government database, called the Central Identities Data Repository (CIDR). Each person’s biometric data, once stored in the CIDR, is linked to a unique ID assigned to the individual. Biometric data of over a billion Indian residents is now stored on the CIDR, and the central database is used to match beneficiary biometrics across the country, at the point of service.
The central government made a conscious design choice to not use smartcards. The Unique Identification Authority of India (UIDAI), a unit established to implement Aadhaar, did not want to restrict the use of biometric authentication to applications requiring card readers (Dholabhai, 2012). A disadvantage to this choice is that the system requires Internet connectivity to match a biometric input with data stored in the CIDR. Technical errors due to poor Internet connectivity are therefore a potential source of exclusion. Despite concerns raised by other government agencies, the UIDAI maintained the view that with the expansion of high-speed mobile networks, Internet connectivity would grow more reliable with time, particularly in rural areas (Zelazny, 2012).
Another design choice for both smartcards and Aadhaar is where to set the target accuracy for biometric authentication. Accuracy is dependent on the quality of the biometric data that is captured when beneficiaries are enrolled. However, it is also possible for an individual’s biometrics to change over time. For example, fingerprints tend to wear off, and eye surgeries can alter iris texture patterns. These can result in technical errors, due to a mismatch between the original biometric records (captured during registration) and those being captured in real time (Ramanathan, 2018; Nigam et al., 2019). Populations such as senior citizens and casual laborers are particularly vulnerable to such authentication errors (Khera, 2019a, b).
One way to improve the match accuracy is to base authentication on multiple biometrics rather on a single biometric. Accuracy levels with a combination of fingerprints and iris scans are an order of magnitude better than using only one or the other (UIDAI, 2012). With multiple biometrics, vulnerable populations whose fingerprints have worn off can still authenticate themselves with iris scans. With each additional biometric, the chances of duplication reduce, increasing the accuracy of the biometric match.
The AP government based its smartcards project on fingerprints (of eight to ten fingers). This may have been due to limitations in the storage capacity of biometric data on smartcards. Aadhaar, on the other hand, is based on a multi-modal system of biometric authentication; in addition to fingerprints, iris scans can also be used. However, authentication through iris scans does depend on an additional hardware component that can read the iris. Without this, Aadhaar-based authentication remains dependent on a single biometric.
3.2 Reforms in Payment Systems and Benefits Delivery
Prior to the introduction of biometric IDs, payments and benefits delivery systems for welfare schemes in India were susceptible to delays and inefficiencies, imposing significant costs for both households and the government. Beneficiaries incurred significant time costs in travelling to bank branches and post offices to access benefits. Governments lacked the ability to track and reconcile transactions in the paper-based system, exacerbating the problem of leakages.
In contrast, biometric IDs linked to bank accounts allow for remote authentication of beneficiaries and the issuance of digital payments. Banks can also hire agents (called business correspondents) to make last-mile payments to beneficiaries using PoS devices that support biometric authentication. This eliminates the need for beneficiaries of the NREGS or the SSP programs to travel to a bank branch or a post office. Instead, payments can be made by business correspondents, at the beneficiary’s doorstep. Figure 20.3 illustrates a sample process flow for a biometrically enabled payment system.
Biometric authentication on PoS devices also allows for digital record-keeping, as digital receipts are generated for each transaction (Muralidharan et al., 2020b). Governments can therefore reconcile each transaction made to beneficiaries. This is useful particularly in the context of the PDS. The digital transaction records created at an FPS allow the government to reconcile household purchases with the subsidized food rations allocated to each shop. If an FPS dealer records transactions of an amount lesser than the entitlements allocated, the government can reduce future allocations, to adjust for the previous month’s balance. This penalty disincentivizes the FPS dealer from over-reporting the volume of goods disbursed to beneficiaries each month. Regular reconciliation can also, in principle, increase government savings by restricting expenditures to only the quantity of commodities required by beneficiaries.
4 Implementation Challenges
As this textbook argues, in order to achieve development impact, basic technologies need to be modified and adapted for local context. The process of building an appropriate biometric ID technology, and then implementing it on a large scale, is not without challenges. As highlighted in the prior section, certain design choices for biometric technology can lead to the exclusion of significant numbers of genuine beneficiaries. In this section, we show how these choices resulted in exclusion errors and how the two state governments in Andhra Pradesh and Jharkhand dealt with challenges in each of the following steps involved in implementing a biometric authentication system.
4.1 Enrollment of Beneficiaries
The most resource-intensive step in any biometric ID program is the enrollment of individuals. In both the AP Smartcards and Jharkhand Aadhaar programs, the government entered into partnerships with private service providers to execute enrollment.
In the case of the AP government, it was difficult to incentivize private players (mainly banks) to participate in the biometric program. Partnership with the AP government required private service providers to bear the upfront cost of infrastructure required for biometric smartcards, in return for a 2% commission on each payment transaction. The government hoped that banks would view this as a business opportunity, enabling them to establish a foothold in rural areas. However, banks were uncertain about the profitability of the program and were therefore reluctant to participate. The banks that did participate most likely did so under pressure from the state government (Mukhopadhyay et al., 2013).
These service providers faced a number of issues while enrolling beneficiaries. There was a lack of coordination with local government officials, which resulted in weak mobilization of beneficiaries at enrollment camps. Technical problems also cropped up, including hardware and software glitches and errors in the lists of eligible beneficiaries supplied by the AP government (ibid.).
Because a substantial share of beneficiaries remained unenrolled, the government was concerned that the high volume of payments to non-enrolled beneficiaries would re-open the door to corruption and leakages. They deployed two strategies in response. First, to incentivize beneficiaries to enroll for smartcards, the government ordered the stoppage of payments to non-enrolled beneficiaries. NREGS and SSP funds were placed in suspended accounts until beneficiaries completed their enrollment. But due to the slow pace of enrollment, the government finally had to relax its stance, allowing manual payments as long as beneficiaries were simultaneously enrolling for biometric IDs. Second, the government tried eliminating commissions for payments that were not biometrically authenticated. The enforcement of this rule required government officials to regularly monitor transaction-level data. However, the government was unable to enforce this rule, as banks were slow to respond to the government’s requests for detailed transaction-level data (ibid.).
Unlike in AP, enrollment in Jharkhand took place at a rapid pace. The government of Jharkhand’s efforts of introducing biometric authentication in the PDS had the advantage of being built on top of the national biometric ID program – Aadhaar which was already being rolled out by the central agency, the UIDAI.
The UIDAI worked through state governments to hire private entities known as enrollment agencies. These had to be certified by the UIDAI to carry out enrollment (Sathe, 2014; Sen, 2019). Contracted agencies set up mobile Aadhaar enrollment centers. At these centers, fingerprints for all ten fingers and iris scans were obtained. Individuals also were required to submit existing identity documents as proof of their demographic details (including name, address, date of birth, and gender). The information collected by the enrollment agency was then encrypted and sent to the UIDAI, which would compare the incoming data with data already stored on the CIDR to check for duplicates. If there were no duplicates, the UIDAI would generate a 12-digit unique ID number (called the Aadhaar number) and issue a card by post, with the individual’s Aadhaar number and demographic details. If there were any discrepancies identified in the deduplication step, the individual would be informed of next steps for remediation (Misra, 2019). The enrollment agencies were paid a commission for every successful enrollment (Zelazny, 2012).
Over the course of a decade, the UIDAI issued over a billion cards to Indian residents. This was aided by the incentive structure the UIDAI had put in place (i.e., a fixed incentive paid for each successful enrollment). The Government of India also made Aadhaar enrollment mandatory for a number of welfare schemes (including the PDS and the NREGS) and made the Aadhaar card a legitimate identity document for a variety of other purposes (Perrigo, 2018).
The Aadhaar incentive structure and rapid pace for enrollments did, however, have an impact on the quality of data collected by enrollment agencies and made enrollment vulnerable to malpractice. Demographic data collected by enrollment agencies were prone to error, since they were dependent on the quality of paper documents submitted. Misspelt names and incorrect dates of birth have been widely reported (Khera, 2019a, b). In addition, mix-ups in biometric data were reported, with the biometrics of one family member mixed together with other family members’ records (Mohammed, 2018).
Ultimately, the Indian government had to blacklist 34,000 enrollment agencies in the period 2010–2017 (Hebbar, 2017). Reasons for dismissal from the program included charging individuals a fee for enrollment (even though enrollment for Aadhaar was supposed to be free) and using weak security protocols that led to leaks in the demographic data of a large number of applicants (Khaira, 2018; Surabhi, 2017). As with AP Smartcards, technical issues in enrollment were also common, especially in remote areas. This made it difficult for individuals in some regions to apply for Aadhaar cards (Matthew, 2019). Those lacking high-quality biometrics and those unable to physically travel to an enrollment agency (owing to disability or old age) found it difficult to enroll (Drolia, 2018).
4.2 Deduplication of Records
To ensure that the same individual is not able to obtain more than one biometric ID, it is important to run regular deduplication checks using algorithms that can accurately identify duplicates. In AP, the government made banks responsible for carrying out deduplication checks. With deduplication, the state hoped to establish a clean beneficiary database of eligible and legitimate beneficiaries for its welfare schemes. However, banks made limited progress in implementing robust deduplication checks. As the development of advanced deduplication software was still in its early stages, banks were at best able to carry out rudimentary checks, such as a village-level text deduplication of beneficiaries (Mukhopadhyay et al., 2013).
In comparison, Aadhaar’s software and hardware architecture enabled superior detection of duplicates. The system uses multiple biometrics (fingerprints and iris scans) for authentication, and the underlying data are stored in a central server called the CIDR. While a biometric database of ten fingerprints can ensure a deduplication accuracy of more than 95%,Footnote 6 with iris scans, the accuracy can increase to more than 99%. However, as noted earlier, the accuracy rates are highly dependent on the quality of data collected during the enrollment stage (UIDAI, 2009).
As deduplication algorithms had never been tested at this scale, the UIDAI decided to select three vendors rather than one. It hoped that competition among the three vendors would yield greater accuracy. The vendors were paid according to the number of deduplication checks they were able to carry out, and the accuracy of the checks determined the allocation of payments across the three vendors (Sharma, 2016). It is not technically feasible for any deduplication algorithm to achieve 100% accuracy, so there is always a possibility of duplicate Aadhaar cards being generated for the same individual (Abraham et al., 2018; Agarwal, 2013). Given the scale at which the program has been rolled out, and challenges with the quality of the enrollment process, accuracy rates of more than 99% for a population of over one billion can still create a substantial number of duplicatesFootnote 7 (Matthews, 2016; Zelazny, 2012).
4.3 Authentication of Identity
Errors in authentication occur when genuine beneficiaries are unable to authenticate themselves using biometric IDs. This can occur for many reasons, and the consequences can be severe: authentication errors without a manual override option (which is discussed in the next section) can strip beneficiaries of their legal entitlements.
In AP, biometric data were stored on the smartcard itself, so authentication could occur offline. However, the system was still prone to authentication errors, mainly due to technical problems with the PoS machines used. In some cases, the PoS reader was unable to recognize fingerprints, despite multiple attempts; in other cases, intermittent lack of power rendered the POS devices useless. Another source of authentication errors was the poor quality of biometric and demographic data collected during enrollment (Mukhopadhyay et al., 2013).
Aadhaar has been vulnerable to some of the same authentication errors experienced in the AP Smartcards program. In addition, it has suffered from poor Internet connectivity in some regions. With biometric data stored on a central server (the CIDR), all real-time authentication requests require connectivity to the Internet. In rural areas with poor Internet connectivity, online authentication proved a major challenge and a cause for significant exclusion of beneficiaries from welfare schemes (Dreze et al., 2017; Sneha, 2017).
The UIDAI did create certain alternatives to online authentication, although their use has been limited. One such alternative is the use of a one-time password (OTP) sent to the mobile number linked to an individual’s UID. However, the use of an OTP naturally requires an active mobile phone number with good mobile connectivity. Early on, mobile phone ownership and connectivity were still lacking in rural areas in India, especially in states with low teledensity such as Jharkhand. Administrative errors in the linkage of mobile numbers to Aadhaar IDs – including the entry of incorrect mobile numbers during enrollment, and the use of “head of household” mobile numbers rather than individuals’ numbers – reduced the reliability of the OTP alternative (Dreze et al., 2017; Muralidharan et al., 2020b). In addition, awareness of this fallback mechanism has been quite low (Abraham etal., 2018).
A second alternative for offline Aadhaar authentication is the use of offline PoS devices. Offline PoS devices capture the beneficiary’s fingerprints but cannot perform real-time authentication. Beneficiaries can still collect their entitlements after scanning their fingerprints. The operators of the offline PoS devices are required to regularly synchronize their transaction logs with the central government server by accessing the Internet. To ensure that logs are regularly synchronized, new transactions are not authorized if the operator has not synchronized earlier logs. To minimize inclusion errors (wherein non-eligible beneficiaries access welfare schemes), the use of PoS devices for offline authentication had been limited to areas with extremely poor connectivity (Muralidharan et al., 2020b).
Another source of authentication errors in Jharkhand was the incomplete or incorrect linking of Aadhaar IDs with existing government beneficiary records, such as that of the PDS. Once beneficiaries of the PDS had obtained an Aadhaar ID, they were required to link their Aadhaar numbers with their existing PDS ration card numbers, a process known as “seeding.” This was a non-trivial process and imposed significant costs on beneficiaries, as they often had to make repeated trips to their local government offices to ensure that ration cards had been seeded. Families with even a single member lacking the seeded ration card were unable to access their PDS benefits. The seeding process was also prone to data entry errors (Dreze et al., 2017; Sneha, 2017).
4.4 Manual Overrides
Given the many challenges described so far, there have been legitimate concerns that making biometric IDs mandatory for public benefits can lead to exclusion of genuine beneficiaries. To reduce exclusion errors, one approach is to create manual overrides. This entails the use of non-biometric authentication, for use when it is impossible to biometrically authenticate an individual. However, manual overrides bring the risk of increased inclusion errors. In the absence of sound protocols for manual overrides, non-eligible beneficiaries can easily access welfare schemes, undermining one of the primary benefits of biometric ID programs.
As discussed earlier, a significant proportion of beneficiaries in AP were never enrolled for biometric smartcards, because the government was unable to incentivize banks and beneficiaries to increase their participation in the smartcard system. This resulted in heavy reliance on manual overrides. Business correspondents continued to make payments (in cash) to unenrolled beneficiaries and to individuals unable to biometrically authenticate themselves. To minimize inclusion errors, the override protocol required business correspondents to authenticate using paper-based ID documents, and for record-keeping, beneficiaries were made to stamp their fingerprints on a pre-filled beneficiary roster (Mukhopadhyay et al., 2013).
In Jharkhand, Aadhaar was in effect a mandatory requirement for accessing welfare schemes such as the PDS; and compared to AP, the use of manual overrides was limited. Offline PoS machines and OTPs were the only fallback options available to those with an Aadhaar ID who could not authenticate biometrically (Dreze et al., 2017). Those without an Aadhaar card (and those whose entire family had failed to “seed” their Aadhaar cards with the ration database) were, in effect, excluded from the PDS. Although the central government did issue notifications to maintain exemption mechanisms, these mechanisms were rarely used (Dutta, 2019).
4.5 Political Buy-In
The successful implementation of any large-scale reform hinges on strong political support. This is often hard to achieve, as government officials and frontline service delivery workers who have benefited under the status quo have little incentive to ensure that reforms are successfully implemented – even when these reforms improve welfare outcomes for society as a whole (Hoff & Stiglitz, 2008; Olson, 1965). Fortunately, AP’s smartcard program had strong political support among senior state government officials. Senior bureaucrats invested significant time and effort to ensure that the smartcards were successfully rolled out. The AP government was also strongly committed to making welfare schemes such as the NREGS and the SSP easily accessible for beneficiaries. This is reflected in the fact that the state had one of the highest fund utilization rates for welfare schemes in the country (Mukhopadhyay et al., 2013).
However, support for the program from local government officials (at the district and sub-district levels) was weak, and this hampered progress in implementing the reform. In addition, business correspondents (BCs) hired by banks to deliver payments to beneficiaries became a sought-after position, which politicized the process of selecting BCs. Politicians would lobby for their own candidates, for political gains. This stalled the selection process in many villages, affecting the delivery of payments. There were also instances of active resistance to the smartcard program from local government officials, and local officials were more likely to convey negative anecdotes about the reform. In fact, opposition from local officials was strong enough for the government to consider abandoning the reform in 2013 (ibid.).
Aadhaar, on the other hand, received strong support early on from important decision-makers in the central government. This helped the program gain political mileage, despite opposition from prominent politicians and external actors related to exclusion risks and privacy concerns (Misra, 2019). Importantly, Aadhaar aligned with the government’s objective of reducing the fiscal burden from leakages in welfare schemes (Government of India, 2015). Indeed, fiscal savings from biometric authentication was the primary benefit highlighted in government reports of Aadhaar pilots in various welfare schemes (Dreze & Khera, 2018; Government of India, 2016). A change in the central government’s ruling party midway through the program’s implementation had little impact on the pace at which the Aadhaar cards were issued to Indian residents and linked with welfare schemes (Sathe, 2014).
State governments that implemented Aadhaar-Based Biometric Authentication (ABBA) in their welfare schemes devoted significant resources to the reform, following the central government’s policy priority of reducing leakages. At the time, state-level reforms were politically popular thanks to strong voter support for reforms addressing corruption and fraud (Sukhtankar & Vaishnav, 2015). As such, the Jharkhand government was willing to institute ABBA in its welfare schemes even if it came at the cost of excluding some genuine beneficiaries (Muralidharan etal., 2020b).
The Jharkhand government’s main policy priority was to increase fiscal savings through ABBA, and this is reflected in the limited use of manual overrides. The state also increased the rate of beneficiary deletion from the PDS rolls shortly before ABBA was introduced. The government claimed that beneficiary names lacking an Aadhaar number were primarily “ghost” beneficiaries. The government also decided to reconcile grain stocks at PDS shops, with the aim of reducing its expenses on the PDS. Reconciliation entailed adjusting the delivery of commodity grains to shops based on earlier transaction records. Reconciliation proved quite unpopular, as FPS dealers generally passed a significant proportion of the government’s reduction in benefits along to beneficiaries. In fact the government had to temporarily suspend and relax its reconciliation protocols in the face of opposition from FPS dealers and beneficiaries (ibid.).
4.6 Innovation and Iteration at Scale
The implementation challenges highlighted here illustrate how biometric ID technology, like any other technological innovation, must be iteratively modified to the context in which it is being deployed. The digital identity solutions deployed in AP and Jharkhand were influenced not only by technology design decisions but also by the broader financial ecosystem and local policy environment. As solutions were scaled across India by UIDAI, the database architecture and hardware became somewhat fixed; but variation in delivery was facilitated through policy and protocol design. Banks needed the right incentives, and state-level agencies needed tools to monitor implementation to ensure that the gains from innovation were reaching beneficiaries. In line with the framework for Development Engineering outlined in this book, the features of the solution were continuously revisited and modified as implementation expanded.
One lesson from this is the importance of researchers engaging with governments and not shying away from the politics of implementation. As this case study demonstrates, the AP and Jharkhand governments weighed the benefits from reduced leakages against the accompanying increase in exclusion errors resulting from biometric authentication. Their decisions were ultimately a function of policy priorities. The AP government’s focus on improving beneficiary experience explains some of the design details of AP Smartcards, including their commitment to offline authentication, manual overrides, and payments to non-enrolled beneficiaries. The Jharkhand government’s focus on fiscal savings explains why Aadhaar was in effect made mandatory for access to welfare services and why limited override mechanisms were made available for beneficiaries.
5 Results from Large-Scale Evaluations
Whether a technological innovation can be viewed as contributing to economic development depends on the rigor with which it is evaluated. A priori, it was unclear whether welfare beneficiaries in India would be better off with biometric IDs since these systems can exclude real beneficiaries. They can be very costly, and faulty implementation can compromise performance. It is important to measure rates of inclusion and exclusion, and other user experiences, to assess the overall welfare impacts of such a large-scale program. In addition, the massive scale of deployment of digital IDs across multiple states of India made evaluation crucial for understanding impacts on the broader economy. Implementation within large geographies can create unique dynamics that are never captured by small-scale evaluations.
Here, we will describe the results of two large-scale randomized evaluations implemented by the Payments and Governance Research Program (Muralidharan et al., 2016, 2020b). What makes these evaluations unique is the scale at which they were conducted, across two varied contexts.
5.1 Experimentation at Scale
The use of RCTs to evaluate the impact of welfare programs has become a common tool in social science research. RCTs involve randomly assigning a treatment (or policy intervention) across a population and then comparing outcomes of those who receive the intervention with those who do not. As treatment and control groups are randomly assigned, RCTs generate credible estimates of the causal impact of an intervention, within the context of the population under study.
There have been concerns about the representativeness of results from RCTs beyond the immediate context of the study sample. If the study’s sample is not representative of the population a government is interested in, which findings can be generalized? How should the evidence from a small trial inform a larger-scale policy intervention? (see footnote 7) Spillover effects – in particular, how the effects of the intervention interact with those who do not receive the intervention – can become pronounced at large scale. An evaluation that does not measure how changes in welfare programs affect the wider economyFootnote 8 may underestimate the impact of an intervention like biometric IDs (Muralidharan & Niehaus, 2017).
The researchers on the team dealt with these issues by carrying out RCTs at scale, which offers the following benefits:
-
1.
The studies consist of samples that are representative of larger populations.
-
2.
The unit of randomizationFootnote 9 is selected such that it is large enough to capture spillover effects.Footnote 10
-
3.
The evaluations can credibly estimate the impact of policies that are implemented by governments at a large scale.
Given the nature of these evaluations, the results can provide valuable input on the impacts of biometric ID systems for policymakers in other developing countries.
5.2 Results from the Evaluation of AP Smartcards
The evaluation of AP’s smartcards initiative was made possible by an unexpected glitch in the program’s roll-out. Because of the challenges in beneficiary enrollment described earlier, there were still pockets in AP where smartcard coverage was limited circa 2009. In eight districts, the banks responsible for enrollment had made little progress, so the project needed to be re-launched in these areas. The re-launch in 2010 provided a unique opportunity to randomize the roll-out of the intervention, enabling the first randomized evaluation of a large-scale biometric ID program (Muralidharan et al., 2016).
The study leveraged the phased introduction of biometric smartcards across 20 million beneficiaries in the eight districts; the results would therefore be representative of a large population. For the study, 112 sub-districts were randomly assigned to receive smartcards immediately (treatment sub-districts), while 45 sub-districts were randomly selected to receive the smartcards 2 years later (control sub-districts). In control sub-districts, the status quo mode of non-biometric authentication continued. By selecting a large unit of randomization (i.e., the sub-district with an average population of 70,000), the study team was able to measure spillover effects of the Smartcards program on broader economic outcomes.
The main outcomes of interest for the study were (1) beneficiary experience and satisfaction with NREGS and SSP and (2) leakage from the programs. Beneficiary experience was captured through household surveys. Leakage was measured as the difference between the dollar value of welfare benefits disbursed by the government (obtained from administrative records) and the value received by beneficiaries, net of transaction costs incurred by beneficiaries in accessing the various welfare schemes (as reported in surveys).
By July 2012, 2 years after the roll-out of the intervention, only 50% of payments were being made via biometric smartcards in treatment sub-districts. However, despite low enrollment, the evaluation found that payments were faster and more predictable – especially for beneficiaries of the NREGS in treatment sub-districts. Beneficiaries in treatment sub-districts spent significantly less time collecting payments, and they received their payments earlier. On the leakage front, the evaluation found an increase in earnings of NREGS and SSP beneficiaries in treated areas. This increase in earnings did not coincide with any major increases in government outlays, suggesting a significant reduction in the leakage of funds. Indeed, leakage in treatment areas was estimated to have reduced by 12.7 percentage points for NREGS beneficiaries and 2.7 percentage points for SSP beneficiaries. These results may be explained by a reduction in government officials pocketing money for beneficiaries who neither worked nor claimed payments. Biometric IDs had made it harder for officials to over-report, since biometric authentication requires beneficiaries to be physically present to verify and receive payment.
Decomposing the effects in treatment areas further, the evaluation found that improvements in payment timeliness were concentrated entirely in villages that had shifted to the new smartcard-linked payment system. In these villages, timeliness improved regardless of whether or not beneficiaries had received biometric smartcards although the reduction in leakages was concentrated entirely among beneficiaries who had received biometric smartcards. These results suggest that the organizational reforms associated with the new payment system – which shifted the responsibility for payments from local government officials over to banks – were a key factor in improving the quality of service delivery and the possession of a biometric ID was key to reducing leakages.
Interestingly, the AP Smartcards program also had a positive impact on the wider economy. In treatment sub-districts, the introduction of smartcards led to broad increases in the earnings of the poor. Most of this came from sources other than the NREGS, suggesting increases in market wages and employment. This helped increase the bargaining power of workers in the labor market, who now had better outside options (Muralidharan et al., 2020a). The evaluation was instrumental in showing that biometric ID systems can have a positive impact both directly – by improving beneficiary experience and reducing leakages – and indirectly, via positive spillover effects on market wages.
5.3 Results from the Evaluation of Aadhaar in Jharkhand
As Aadhaar was rolled out across India, Jharkhand (and its state PDS program) became one of the frontrunners in completing the prerequisites for participation. After being satisfied with the results from small-scale pilot deployments, the Jharkhand government scaled up biometric authentication to ten districts. To evaluate the impacts of the reform, Jharkhand’s government agreed to randomize the ordering of the roll-out across 15 million beneficiaries in the 10 districts. This was after the results of the AP experiment had been shared, building interest of the government in collaborating with the research team (Muralidharan et al., 2020b).
Similar to AP, the roll-out was randomized at the level of a large administrative unit, to capture spillover effects. In the experiment, 132 sub-districts were randomly selected to receive ePoS devices supporting ABBA, of which 87 sub-districts (treatment) would receive the ePoS devices first and the remaining 45 sub-districts (control) would receive the ePoS devices at a later stage.
The first phase of the study consisted of introducing ABBA into the PDS treatment group, without any reconciliation based on FPS transaction data. Despite Jharkhand being a relatively low capacity state, by the time of the follow-up surveys, 96% of the shops in treatment areas had introduced ePoS devices, and 91% reported regularly using the devices to biometrically authenticate and record transactions.Footnote 11
The evaluation results demonstrated a limited impact of ABBA on leakage: the value of food entitlements received by beneficiaries did not vary significantly between treatment and control groups. In addition, the reform did not decrease government spend on food entitlements in treatment areas, suggesting the reform did not reduce diversion of goods to the black market. However, the use of ABBA came at a cost to beneficiaries. Many beneficiaries made multiple unsuccessful trips to the FPS before they were able to authenticate and access food entitlements, resulting in a 17% increase in transaction costs for those in treatment sub-districts. The reform also had a significant negative impact on the value of food entitlements for certain subgroups. For households with a single member failing to seed their Aadhaar card with a ration card, the reform led to a significant drop in the value of rice and wheat received and increased the probability of receiving no commodities at all.
Shortly after ABBA was introduced in the control districts, the state government introduced stricter reconciliation protocols. This “second phase” of intervention involved the government adjusting downward the amount of grain disbursed to each FPS, reflecting stocks that the FPS should have maintained (based on digitized transaction records from prior months). Reconciliation was introduced at the same time in both control and treatment sub-districts, but treatment areas had ePoS devices for a longer period of time (11 months) than control sub-districts (2 months). So in treatment areas, 11 months’ worth of commodities diverted away from beneficiaries would have been adjusted for at the introduction of reconciliation. The analysis suggests that stricter reconciliation of commodities did reduce leakage.
In treatment sub-districts, reconciliation led to a sharper decrease in the value of food grains disbursed by the government, relative to control sub-districts. It also coincided with a sharp reduction in the value of goods received by households in treatment sub-districts. According to the reconciliation protocols used by the government, since FPS dealers in treatment areas had ePoS devices for longer, they were more likely to have undisbursed food grains captured in transaction records. But in practice, it was more likely that FPS dealers had already diverted the undisbursed grains to the open market. This explains why there was a sharper decrease in grains disbursed to households in the treatment sub-districts after reconciliation.
Had the government decided not to hold FPS dealers accountable for past diversions – and had the government given all dealers a “fresh start” – a more favorable outcome may have been obtained. Starting reconciliation on a “clean slate” basis (rather than basing disbursements on past transaction records) would have achieved a more modest reduction in leakage, with minimal impact on the value received by beneficiaries. These results show that the introduction of biometric IDs in Jharkhand, on its own, did not reduce leakage. But combined with reconciliation, there was a reduction in leakage that coincided with a sharp reduction in the value of benefits received by beneficiaries of the PDS.
5.4 Summary
In both states, the introduction of biometric IDs coupled with other reforms (such as reconciliation in Jharkhand) did lead to a substantial reduction in leakage of benefits. However, in AP, the benefits from the reduction of leakage were passed onto beneficiaries, whereas in Jharkhand, the main policy priority was to increase fiscal savings, which ultimately resulted in beneficiary loss. The policy implication is that while biometric IDs can bring down leakages, differences in program and technology design can have an important bearing on whether beneficiaries gain or lose.
6 Conclusions
The studies discussed in this chapter demonstrate that technological interventions like biometric IDs have the potential to increase state capacity and improve last-mile service delivery in developing countries. However, differences in policy priorities and the details of solution design influence the extent to which households benefit from the digitization of welfare schemes.
Results from two large-scale evaluations find that more accurate biometric ID systems, coupled with payments and policy reforms, reduced leakages in welfare schemes in two different states in India. However, there were varying results. In Jharkhand, reduced fiscal leakage came at the expense of excluding genuine beneficiaries who were unable to meet new standards for identification. Exclusion of beneficiaries was low in Andhra Pradesh, where the government was more focused on improving beneficiary experience with welfare programs.
Differences in policy priorities of the two states are reflected in differences in the design of the two ID programs. Manual overrides were available for those unable to biometrically authenticate themselves in AP, but limited overrides were made available to beneficiaries of the PDS in Jharkhand. The use of past transaction records to reconcile grain stocks, combined with the deletion of ration cards from old beneficiary rolls, underlines the Jharkhand government’s priority of fiscal savings. These policies are arguably legitimate political choices. ABBA was introduced into Jharkhand at a time when there was a strong public movement in India against corruption and there was a push from the central government to roll out ABBA in social service delivery. Despite reports of exclusion and preventable starvation deaths, surveys of beneficiaries indicated that while views were polarized,Footnote 12 there was still strong support for Aadhaar reforms to the PDS (Muralidharan et al., 2020b). This highlights that reforms that are harmful to a significant minority can still be politically viable, if they are perceived as being in the larger public interest.
The evaluations also highlight the importance of incorporating regular data collection on beneficiary experience as part of any large-scale technology reform. For example, in AP with access to only administrative data, one may have mistakenly concluded that there was no reduction in leakage, as there was no change in government expenditure. Households’ surveys showed otherwise. Conversely, in Jharkhand, one might conclude that there was a sharp reduction in leakage due to the reduction in government expenditure post-reconciliation. As we now know, Jharkhand’s reduction in government expenditure post-reconciliation came at the cost of excluding eligible beneficiaries and reducing their overall benefits.
Ultimately, the design of technologies for public service delivery is highly political. There are many non-technical failures in the performance of digital identity solutions, and these expose the governance challenges underlying any public sector technology solution.
Discussion Questions
-
1.
Is there a trade-off between inclusion (which leads to leakage) and exclusion errors? If yes, how should governments weigh the trade-off between inclusion and exclusion errors?
-
2.
In addition to the manual overrides discussed in this chapter, are there any other technical/non-technical solutions to the problem of exclusion errors?
-
3.
What role can civil society play in protecting citizens from potential misuse of biometric data?
-
4.
Should the Government of Jharkhand have introduced a “clean slate” reconciliation of PDS stocks, to protect beneficiaries from loss of benefits? Could they have known, ex ante, that it would have disadvantaged beneficiaries?
-
5.
The AP Smartcards program has since been replaced with Aadhaar. Was the state government wise to experiment with a local, offline smartcards solution, given the high costs of transitioning to the central government’s Aadhaar technology?
Notes
- 1.
Taking cognizance of these issues, the Kenyan High Court had in fact temporarily suspended the implementation of the Huduma Namba until adequate legal safeguards were put in place. On the other hand, the Supreme Court in India in a majority opinion (with one judge dissenting) allowed the use of the Aadhaar biometric ID program for welfare schemes, as it was of the opinion that it collected minimal data for a larger purpose, i.e., to provide services to the poor.
- 2.
The Payments and Governance Research Program is a team of researchers led by Karthik Muralidharan, Paul Niehaus, and Sandip Sukhtankar based at the Jameel Poverty Action Lab (J-PAL), South Asia (see https://sites.google.com/ucsd.edu/pgrp/home).
- 3.
To illustrate how over-reporting and underpayment took place in practice, consider a worker who has genuinely earned Rs. 100; the Field Assistant might report in the muster rolls that he is actually owed Rs. 150 but pay the worker only Rs. 90 thereby pocketing Rs. 50 through over-reporting and Rs. 10 through underpayment, resulting in a leakage rate of 40% (Muralidharan et al., 2016).
- 4.
Old age (>65), widow, disabled, or certain displaced traditional occupations.
- 5.
Another (more costly) option is combining both designs with the use of a chip-based smartcard with a unique ID linked to a central server. This allows for both online and offline authentication. National biometric ID cards in Indonesia and the Philippines utilize this design (Asian Development Bank, 2016).
- 6.
This is referred to as the true acceptance rate (TAR) of a biometric system and is a measure of the number of times a biometric system can correctly verify an individual’s claim of identity. For example, a TAR of 95% implies that if 10% of individuals in a population of 1000 try to register for a biometric ID twice, the system would incorrectly generate duplicate biometric IDs for 5 individuals. Similarly, for a population of 1.2 billion, this would imply duplicate biometric IDs for 6 million individuals. An increase in the TAR to 99% brings this number down to 1.2 million individuals.
- 7.
The results of an intervention implemented by an NGO at a small scale, for example, may be very different if the same intervention is rolled out by a state/provincial government at a much larger scale.
- 8.
For example, as we show later, an improvement in access to the NREGS forced landowners in Andhra Pradesh to increase the wages for laborers as the laborers now had better outside options.
- 9.
The unit of randomization essentially refers to the level at which the treatment/policy is introduced. For example, the policy may be introduced randomly across individuals in which case the unit of randomization would be at the individual level, or it may be introduced randomly across sub-districts in which case the unit of randomization is at the sub-district level.
- 10.
For example, if the unit of randomization is the village or the sub-district, then researchers can capture spillovers at the village or sub-district level.
- 11.
This difference was mainly because enrollment rates for Aadhaar cards were already quite high prior to it being introduced in Jharkhand’s PDS. The central government had made high Aadhaar coverage and ABBA for welfare schemes a nationwide priority. In AP, there was no supporting infrastructure such as Aadhaar that was available to the state government (Muralidharan et al., 2020b).
- 12.
The ruling government in Jharkhand at the time ABBA was introduced in fact lost its majority in the next elections.
References
Abraham, R., Bennett, E. S., Bhushal, R., Dubey, S., Li, Q., Pattanayak, A., & Shah, N. B. (2018). State of Aadhaar report 2017–18. Technical report, IDinsight.
Agarwal, S. (2013, March 5). Duplicate Aadhaar numbers within estimates: UIDAI. Mint.
Alkire, S., & Seth, S. (2013). Selecting a targeting method to identify BPL households in India. Social Indicators Research, 112(2), 417–446. https://doi.org/10.1007/s11205-013-0254-6
Asian Development Bank. (2016). Identity for development in Asia and the Pacific. Asian Development Bank.
Banerjee, S., & Sharma, S. (2018). An offline alternative for Aadhaar-based biometric authentication. Ideas for India. https://www.ideasforindia.in/topics/productivity-innovation/an-offline-alternative-for-aadhaar-based-biometric-authentication.html
Bardhan, P. (1997). Corruption and development: A review of issues. Journal of Economic Literature, 35(3), 1320–1346.
Bhatia, R. (2018, October 5). The Indian Government’s astonishing hunger for citizen data. The New York Times.
Census of India. (2003). Project review: Multi-purpose national identity card. eCensus India. Issue 17.
Dahir, A. L. (2020, January 28). Kenya’s new digital IDs may exclude millions of minorities. The New York Times.
Dholabhai, N. (2012, January 8). Unique ID better than PC smart card: Montek. The Telegraph.
Dreze, J. (2016, March 15). The Aadhaar Coup. The Hindu.
Dreze, J., Khalid, N., Khera, R., & Somanchi, A. (2017). Aadhaar and food security in Jharkhand pain without gain? Economic & Political Weekly, 52(50), 50–59.
Dreze, J., & Khera, R. (2015). Understanding leakages in the public distribution system. Economic & Political Weekly, 50(7), 39–42.
Dreze, J., & Khera, R. (2018, February 8). Aadhaar’s $11-bn question. The Economic Times.
Drolia, R. (2018, February 15). 81 year old seeks exemption for super elderly. The Times of India.
Dutta, P. (2008). The performance of social pensions in India: The case of Rajasthan. SP Discussion Paper No. 0834. World Bank.
Dutta, P., Murgai, R., & Howes, S. (2010). Small but effective: India’s targeted unconditional cash transfers. Economic & Political Weekly, 45(52), 63–70.
Dutta, S. (2019, February 8). Refusing to learn from tragedy, Jharkhand Demands Aadhaar for rations. The Wire.
Ferraz, C., Finan, F., & Moreira, D. B. (2012). Corrupting learning. Journal of Public Economics, 96(9–10), 712–726. https://doi.org/10.1016/j.jpubeco.2012.05.012
Gelb, A., & Metz, D. A. (2018). Identification revolution: Can digital ID be harnessed for development? Center for Global Development.
Global Hunger Index Scores by 2020 GHI Rank. https://www.globalhungerindex.org/ranking.html. Accessed 28 Dec 2020.
Goldberg, J., Gine, X., & Yang, D. (2010). Identification strategy: A field experiment on dynamic incentives in rural credit markets. Policy Research Working Papers. https://doi.org/10.1596/1813-9450-5438
Government of India. (2001). Report of the Group of Ministers on National Security.
Government of India, Ministry of Finance. (2015). Economic Survey 2014–15.
Government of India, Ministry of Finance. (2016). Economic Survey 2015–16.
Government of India, Ministry of Finance. (2020). Economic Survey 2019–20.
Hebbar, N. (2017, April 12). Aadhaar robust; the poor have no complaints about it: Ravi Shankar Prasad. The Hindu.
Hoff, K., & Stiglitz, J. E. (2008). Exiting a lawless state. The Economic Journal, 118(531), 1474–1497. https://doi.org/10.1111/j.1468-0297.2008.02177.x
Khaira, R. (2018, January 3). Rs 500, 10 minutes, and you have access to billion Aadhaar details. The Tribune.
Khera, R. (2019a, March 10). The many pitfalls of Aadhaar. The Hindu Business Line.
Khera, R. (2019b). Aadhaar failures: A tragedy of errors. Economic and Political Weekly. https://www.epw.in/engage/article/aadhaar-failures-food-services-welfare
Matthew, A. (2019, October 21). Aadhaar failure: Several thousands in three districts in Odisha denied ration for two months. National Herald.
Matthews, H. V. (2016). Flaws in the UIDAI process. Economic & Political Weekly, 51(9), 74–78.
Misra, P. (2019). Lessons from Aadhaar: Analog aspects of digital governance shouldn’t be overlooked. Pathways for Prosperity Commission.
Mohammed, A. (2018, July 26). Biometric mix-up affects nearly 2 crore Aadhaar holders. The New Indian Express.
Mookherjee, D. (2014). MNREGA: Populist leaky bucket or successful anti-poverty Programme? Ideas for India. https://www.ideasforindia.in/topics/poverty-inequality/mnrega-populist-leaky-bucket-or-successful-anti-poverty-programme.html
Mukhopadhyay, P., Muralidharan, K., Niehaus, P., & Sukhtankar, S. (2013). Implementing a biometric payment system: The Andhra Pradesh experience. Technical Report. University of California, San Diego.
Muralidharan, K., Das, J., Holla, A., & Mohpal, A. (2017). The fiscal cost of weak governance: Evidence from teacher absence in India. Journal of Public Economics, 145, 116–135. https://doi.org/10.1016/j.jpubeco.2016.11.005
Muralidharan, K., & Niehaus, P. (2017). Experimentation at scale. Journal of Economic Perspectives, 31(4), 103–124. https://doi.org/10.1257/jep.31.4.103
Muralidharan, K., Niehaus, P., & Sukhtankar, S. (2016). Building state capacity: Evidence from biometric smartcards in India. American Economic Review, 106(10), 2895–2929. https://doi.org/10.1257/aer.20141346
Muralidharan, K., Niehaus, P., & Sukhtankar, S. (2020a, March). General equilibrium effects of (improving) public employment programs: Experimental evidence from India. Working paper 23838, National Bureau of Economic Research.
Muralidharan, K., Niehaus, P., & Sukhtankar, S. (2020b). Identity verification standards in welfare programs: Experimental evidence from India. https://doi.org/10.3386/w26744.
Niehaus, P., Atanassova, A., Bertrand, M., & Mullainathan, S. (2013). Targeting with agents. American Economic Journal: Economic Policy, 5(1), 206–238. https://doi.org/10.1257/pol.5.1.206
Niehaus, P., & Sukhtankar, S. (2013). The marginal rate of corruption in public programs: Evidence from India. Journal of Public Economics, 104, 52–64. https://doi.org/10.1016/j.jpubeco.2013.05.001
Nigam, I., Keshari, R., Vatsa, M., Singh, R., & Bowyer, K. (2019). Phacoemulsification cataract surgery affects the discriminative capacity of iris pattern recognition. Scientific Reports, 9(1). https://doi.org/10.1038/s41598-019-47222-4
Olken, B. A. (2006). Corruption and the costs of redistribution: Micro evidence from Indonesia. Journal of Public Economics, 90(4–5), 853–870. https://doi.org/10.1016/j.jpubeco.2005.05.004
Olken, B. A., & Pande, R. (2012). Corruption in developing countries. Annual Review of Economics, 4(1), 479–509. https://doi.org/10.1146/annurev-economics-080511-110917
Olson, M. (1965). Logic of collective action: Public goods and the theory of groups (Harvard economic studies. v. 124). Harvard University Press.
Pai, S. (2013, December 29). Delayed NREGA payments drive workers to suicide. Hindustan Times.
Perrigo, B. (2018, September 28). India has been collecting eye scans and fingerprint records from every citizen. Here’s what to know. Time.
Pritchett, L. (2009). Is India a flailing state?: Detours on the four lane highway to modernization. HKS Faculty Research Working Paper Series RWP09-013.
Ramanathan, U. (2018, September 29). More equal than others. The Indian Express.
Reinikka, R., & Svensson, J. (2001). Explaining leakage of public funds. Policy Research Working Papers. https://doi.org/10.1596/1813-9450-2709.
Sathe, V. (2014). Managing massive change: India’s Aadhaar, the world’s most ambitious ID project (innovations case narrative: Project Aadhaar). Innovations: Technology, Governance, Globalization, 9(1–2), 85–111. https://doi.org/10.1162/inov_a_00204
Sen, S. (2019). A decade of Aadhaar: Lessons in implementing a foundational ID system. ORF Issue Brief No. 292. Observer Research Foundation.
Sharma, R. S. (2016). UIDAI’s public policy innovations. NIPFP Working Paper Series No. 176. National Institute of Public Finance and Policy, New Delhi.
Singh, S. (2020, February 9). Once upon a time, a national ID project found: Complicated to prove citizenship. The Indian Express.
Sneha. (2017). Aadhaar-based biometric authentication for PDS and food security: Observations on implementation in Jharkhand’s Ranchi District. Indian Journal of Human Development, 11(3), 387–401. https://doi.org/10.1177/0973703017748384
Sukhtankar, S., & Vaishnav, M. (2015). Corruption in India: Bridging research evidence and policy options. India policy forum. National Council of Applied Economic Research, 11(1), 193–276.
Surabhi. (2017, January 9). Touts cashing in on Aadhaar card rush. The Hindu Business Line.
UNICEF, et al. (2020). Levels and trends in child mortality. UNICEF.
Unique Identification Authority of India. (2009). Biometrics design standards for UID applications. UIAI.
Unique Identification Authority of India. (2012). Role of biometric technology in Aadhaar enrollment. UIAI.
United Nations Development Programme. (2020). Human development report 2020. UNDP.
World Bank Group. (2003). World development report 2004: Making services work for poor people. World Bank.
World Bank Group. (2019a). Identification for development (ID4D) 2019 annual report (English). http://documents.worldbank.org/curated/en/566431581578116247/Identification-for-Development-ID4D-2019-Annual-Report
World Bank Group. (2019b). Identification for development (ID4D) practitioner’s guide. http://documents1.worldbank.org/curated/en/248371559325561562/pdf/ID4D-Practitioner-s-Guide.pdf
Xavier, G., Goldberg, J., Sankaranarayanan, S., Sheerin, P., & Yang, D. (2012). Use of biometric technology in developing countries. In R. Cull, A. DemirgucKunt, & J. Morduch (Eds.), Banking the world: Empirical foundations of financial inclusion. MIT Press.
Zelazny, F. (2012). The evolution of India’s UID program: Lessons learned and implications for other developing countries (CGD policy paper 008). Center for Global Development. http://www.cgdev.org/content/publications/detail/1426371
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.
The images or other third party material in this chapter are included in the chapter's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the chapter's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
Copyright information
© 2023 The Author(s)
About this chapter
Cite this chapter
Nair, A., Eskici, B. (2023). Digital Public Services: The Development of Biometric Authentication in India. In: Madon, T., Gadgil, A.J., Anderson, R., Casaburi, L., Lee, K., Rezaee, A. (eds) Introduction to Development Engineering. Springer, Cham. https://doi.org/10.1007/978-3-030-86065-3_20
Download citation
DOI: https://doi.org/10.1007/978-3-030-86065-3_20
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-86064-6
Online ISBN: 978-3-030-86065-3
eBook Packages: Earth and Environmental ScienceEarth and Environmental Science (R0)