Decentralised Targeting of Transfer Programmes a Teassessment

Decentralised targeting of transfer programmes: a reassessment Summary Decentralised governance has been widely adopted in developing countries in the hope of incorporating local information into policymaking, enhancing accountability and encouraging democratic participation in the delivery of public services to the poor and needy. However, evaluations of experience with this change have highlighted problems of corruption, elite capture, and clientelism that have undermined the success of decentralisation in improving targeting of transfer programmes. Given recent advances in information technology, this chapter suggests the need to consider suitable reforms, including enhanced monitoring and recentralisation initiatives that reduce local officials’ scope for discretion. It provides an overview of recent research on these topics, and discusses key questions raised by their findings. The period between 1950 and 1990 was characterised by centralised implementation of public benefit programmes in developing countries, whereby ‘implementation’ of a benefit programme refers to its management and allocation across potential recipients. A wide range of types of benefits were involved, including land, water, and subsidised farm inputs such as credit, fertiliser, and seeds; local infrastructures, such as roads, canals, sanitation, and public health; and workfare programmes and welfare services, such as low-income housing, food aid, pensions, education, and health benefits. Top-down, centralised implementation during this early phase involved the delegation of these tasks to a bureaucracy appointed by and accountable to a central government at either the federal or state level. This system gave rise to growing disenchantment owing to targeting failures, leakages, losses, corruption, and lack of responsiveness to local needs. Subsequently, the past three decades have witnessed a shift from centralised towards decentralised implementation, with authority moving from state bureaucrats to local government officials elected by local citizens. The primary motivation was both to improve the information base of allocation decisions and to align the incentives of officials more with the interests of local citizens. As vividly described in a comparative case study of irrigation management systems in South Korea and the Indian state of Andhra Pradesh (Wade 1996), decentralisation was expected to achieve these objectives owing to the closer proximity and accountability of local delivery officials to the citizens that were meant to be served. The 2004 World Development Report accordingly chose decentralised delivery of public benefits as its theme and endorsed it with a sense of hope, noting that: Too often services fail poor people in access, in quality, and in affordability … this year’s World Development Report argues that services can be improved by putting poor people at the center of service. How? By enabling the poor to monitor and discipline service providers, by amplifying their voice in policymaking, and by strengthening the incentives for providers to serve the poor. (World Bank 2004) However, the argument that decentralisation enhances accountability of service providers remains highly controversial. Historically influential counterarguments were made by designers of both the US and Indian constitutions that local governments are more prone than central governments to ‘capture’ by local elites, especially in areas of high inequality, poverty, and lack of popular participation in politics (Hamilton, Madison, and Jay 1787; Pal 2019). These objections are complemented by additional concerns that local democracy in developing countries tends to be characterised by higher levels of political clientelism, where local incumbents manipulate benefit allocations to benefit their loyal supporters or swing voters and so increase their chances of being re-elected. Such problems create non-trivial trade-offs between centralised and decentralised implementation, which have been the subject of extensive research in recent years. Theoretical analyses have argued that the overall outcomes are likely to be highly context-dependent (Bardhan and Mookherjee 2000, 2005, 2006a). This point appears to be borne out by a large body of empirical research (reviewed in Mansuri and Rao 2013 and Mookherjee 2015) evaluating the functioning of local governance in various developing countries. A growing awareness of these problems has motivated some recent reform efforts, especially those involving enhanced monitoring or reducing the extent of discretion provided to local government officials. This form of ‘recentralisation’ has often drawn on emerging new technology and the ‘big data’ capabilities of central governments. The latter part of this chapter provides a perspective on the potential benefits, drawbacks, and wider implications of these recentralisation initiatives. Some qualifications are in order. First, ‘recentralisation’ initiatives are confined to the administration of transfer programmes delivering individually consumed or ‘private good’ benefits, rather than covering infrastructure programmes providing local public goods. They amount to a shift of government expenditure systems in developing countries closer to that of developed countries, where most transfer programmes (for example, social security) are centralised and formula-bound, while the role of local governments is limited to the provision of infrastructure and local public goods. Second, while the ‘first-generation’ literature focused on problems of inter-jurisdictional externalities, scale economies, and taste heterogeneity across jurisdictions, by contrast I focus here on the implications of recentralisation on problems of misallocation owing to governance and accountability defects that have been the topic of the ‘second-generation’ literature on fiscal federalism. 1 While the recent literature has focused mainly on intra-jurisdiction misallocation from capture and clientelism, it has also devoted some attention to related forms of inter-jurisdiction misallocation. But issues of inter-jurisdiction externalities have been largely ignored. However, on a first approximation this omission is not particularly glaring because recentralisation reforms chiefly concern the allocation of transfer benefit programmes that involve negligible externalities (beyond receiving individuals or households) and that benefit from considerable scale economies. The first section of this chapter extends the earlier literature reviews mentioned above to include more recent literature on intra-community targeting distortions, that is, the extent to which local governments succeed in targeting public benefits to intended beneficiaries within their own jurisdictions. The key mechanisms that may prevent successful targeting are elite capture, corruption, and clientelism. Section 3.2 turns to inter-community targeting distortions, resulting from the opportunistic manipulations of programme budgets across different local governments by officials at higher levels. Section 3.3 describes the outcomes of various (‘recentralisation’) reforms intended to reduce these distortions, including attempts to improve monitoring and supervision, institutional alternatives to political decentralisation, and transitioning to formula-bound programmes that reduce the authority of locally elected officials. The evidence suggests that drastic reforms involving elimination of local control over cash and in-kind transfer programmes may be needed to achieve significant improvements in pro-poor targeting. Finally, Section 3.4 discusses other considerations that need to be included in evaluating these reforms, notably the implications for insurance, administrative challenges, and some wider implications for federalism and democracy. 3.1 Intra-community targeting distortions Figure 3.1 depicts the allocation of transfers across different households by a local government within a given community – a ‘private good’ benefit programme targeted for ‘deserving’ households. ‘Deservingness’ refers to a household attribute that is not observed by the central government or in the public domain. In the case of a welfare programme, poor and needy households constitute deserving households. However, for an input subsidy programme intended to promote growth, those deserving are productive households. Local government officials can identify deserving households within the community and have delegated authority over intra-community allocation of a programme budget assigned to them by higher-level governments. Despite being well-informed, local officials may not be motivated to allocate benefits to the needy or otherwise ‘deserving’ households owing to problems of capture, corruption or (political) clientelism. Intra-community misallocation refers to the resulting deviation of actual allocations from those intended. Figure 3.1: Intra-community misallocation The hierarchy within local communities in developing countries is headed by traditional elites (landlords, influential families and notables, tribal or religious leaders, wealthy business interests) and political elites (local government officials). Traditional elites have long-standing social and economic relations with non-elite households, often resembling vertical patron–client ties. Political elites are of a more recent origin, either appointed or elected by local citizens, and subject to periodic turnover. Elite capture refers to symbiotic relationships between traditional and political elites within the community involving the exchange of government benefits or other privileges for financial contributions in the form of campaign contributions or bribes. Corruption consists of the self-diversion of benefits by political elites, or favourable treatment given to households paying bribes or those with personal social connections. Political clientelism refers to political patron–client ties between incumbent political elites and pro-incumbent non-elite households, involving voting for the former in exchange for benefits. In some cases, traditional elites also constitute or overlap with the political elite or have close social connections with the latter (see Chapter 7). In that case, elite capture collapses into a form of corruption. More generally, the distinction between elite capture and corruption is not sharp. For this reason, we club the two phenomena together under a common ‘elite capture’ umbrella. On the other hand, there is a clear conceptual distinction between elite capture and political clientelism, in terms of both their underlying mechanisms and consequences. Elite capture biases benefit allocations in favour of traditional elites vis-a-vis non-elite households (a form of vertical inequity), owing to horizontal social ties or transactions (campaign contributions or bribes). Political clientelism biases allocations across non-elite households in favour of pro-incumbent households (a form of horizontal inequity) motivated by vertical political transactions (exchange of benefits for votes). The vertical equity consequences of clientelism are ambiguous, depending on how pro-incumbent partisanship is correlated with household economic status. For instance, it may enhance vertical equity if poorer households are more prone to support the incumbent in clientelist fashion. While elite capture adversely impacts all non-elite households, the impact of political clientelism is typically favourable for the incumbents’ non-elite supporters (at least in the short run) and unfavourable for other non-elites. 2 Elite capture Most of the earlier (pre-2015) literature focused on elite capture. More recent literature has provided further vivid demonstrations of elite capture, drawing on fine-grained details of the institutional context, besides better data and creative identification strategies. The specific type of local elite varies with the context. In African countries such as Sierra Leone (Acemoglu, Reed, and Robinson 2014) and Malawi (Basurto, Dupas, and Robinson 2019), they are tribal chiefs. In the Indian state of Maharashtra the local elites are large landowners (Anderson, Francois, and Kotwal 2015). In Pakistan the elites are religious leaders (Mehmood and Seror 2021) or long-entrenched political dynastic families (Malik, Mirza, and Platteau 2021). They are political elites in Thailand’s credit programme (Vera-Cossio 2021), and agents appointed by political elites in Uganda’s agricultural extension programme (Bandiera et al. 2020). The relation between traditional elites and non-elites resembles a patron–client relationship, in which clients have traditionally depended on their patrons for employment or insurance against idiosyncratic shocks. Elite capture is then driven by traditional forms of social or economic clientelism, to be contrasted with political clientelism. For instance, in the Maharashtra context (studied by Anderson, Francois, and Kotwal 2015), elite capture took the form of traditional landowning elites suppressing the take-up of a government employment guarantee programme by the poor, in order to keep market wages low and preserve traditional dependency patterns. The extent of elite capture (measured by adverse consequences for non-elites) has typically been higher in communities characterised by greater concentration of power among traditional elites: for example, more unequal land distribution in the case of Indian landed elites, or less political competition between tribal chiefs in Sierra Leone. In both these contexts, village surveys showed greater trust expressed by clients in their patrons in these communities, possibly explained by the greater extent to which people relied on patron support in times of need. While these patterns recur across many different contexts, there are some notable exceptions. In Indonesia (Alatas et al. 2012) or the Indian state of West Bengal (Bardhan and Mookherjee 2006b; Mookherjee and Nath 2020), the extent of elite capture seems quite limited. In Pakistan, Malik, Mirza, and Platteau (2021) found more harmful impacts of election of entrenched dynasts in jurisdictions where they were exposed to greater political competition. The authors explain this via a clientelistic mechanism aggravated by greater political competition, rather than by elite capture (which competition tends to limit). Political clientelism The recent literature has presented growing evidence of political clientelism, where local government officials condition the distribution of private benefits to recipients voting for or expressing political support for their own party (Dunning et al. 2013; Hicken 2011). Citizens are then effectively coerced to vote for the incumbent, reducing the accountability pressure on incumbents and cementing their grip on political power. Benefits are targeted along partisan lines (to favour loyal supporters or swing voters). The type of misallocation resulting from political clientelism differs in essential ways from those resulting from elite capture. Clientelism is often consistent with pro-poor targeting since the poor value government benefits more (at the margin) and are therefore more willing to ‘sell’ their vote in exchange for these benefits (Stokes 2005). So the associated misallocation is different: it need not consist of errors of inclusion, that is, diversion to recipients who are not entitled to receive them, or who are less deserving in terms of need. Within the set of intended beneficiaries, incumbents often discriminate on the basis of political partisanship: rewarding supporters and denying others (especially those that support their political rivals). This results in systematic exclusion errors (sections of the deserving poor that are systematically denied benefits on account of their lack of political support) and associated violations of horizontal equity (where selection among equally deserving groups is based on political partisanship). These forms of misallocation can be hard to detect when programme budget constraints prevent all deserving beneficiaries from receiving benefits, which is often the case in poor countries. Local officials then have discretion over who among the intended beneficiaries will actually receive benefits. Discrimination in favour of their political supporters is frequently difficult to establish by external researchers or auditors. Another difference from elite capture is that clientelism may not require high levels of social or economic inequality within the community, though it usually thrives where poverty is widespread. Greater political competition can enlarge clientelistic distortions, by raising the incentives for insecure incumbents to engage in political favouritism that ends up holding back long-term development. Indeed, this is essentially the explanation provided by Malik Mirza, and Platteau (2021) for their finding in Pakistan that adverse developmental impacts resulted from closer election races that entrenched dynasts won. Compared with capture, clientelism generates other distinct forms of misallocation. It creates a bias among officials in favour of ‘private good’ benefits programmes (especially those of a short-term recurring nature), relative to local public goods (since these are worthless as a clientelistic instrument owing to their non-exclusionary nature). Some authors argue that clientelism generates a broader lack of accountability of elected officials, insofar as residents may be compelled to vote for them despite glaring lapses of governance (for example, manifest mismanagement, neglect of urgent public interest causes, or immoral behaviour). Many recent empirical illustrations of clientelism rely on evidence of these kinds of distortions, associated with the supply of local public goods. For instance, Khemani (2015) found that reported levels of vote-buying in Philippine villages were negatively correlated with health service delivery. Using a regression discontinuity associated with the roll-out of the non-clientelistic programme Bolsa Familia in Brazil, Frey (2019) showed that it was associated with a marked rise in share of health and education shares of municipal government budgets. Using instrumental variable and difference-in-difference methods, Bardhan et al. (2020) provided evidence that voters in West Bengal in India responded to receiving ‘private good’ welfare benefits (employment in workfare programmes, or low-income housing benefits) but not of ‘public good’ benefits (such as local roads and irrigation projects). Consistent with these voting patterns, changes in political competition motivated upper-level officials to discriminate more in favour of local areas controlled by their political party, but this effect was observed only for private good benefits. 3.2 Inter-community misallocation: upper-level manipulation Programme budgets flow down the administrative hierarchy of higher-tier and local governments, as shown in Figure 3.2. The central or provincial government decides on a budget for each district government, which then allocates it between different local community governments at the bottom tier. District government officials may be granted autonomy over the allocation across local governments under their jurisdiction, owing to their superior information concerning relative deservingness of different local communities relative to the provincial administration. However, these officials may abuse their authority by biasing allocations in favour of certain local areas. Figure 3.2: Inter-community allocation Upper-tier officials could be personally biased in favour of either their own home area or areas dominated by their own ethnic group. Alternatively, a partisan bias could be at work in favour of areas where larger programme allocations would help generate more votes (or more salient vote shifts) for their particular party. These distortions could and frequently do arise even in centralised or non-democratic systems. For instance, Burgess et al. (2015) demonstrated that, in periods when strong autocracy prevailed, strong ethnic favouritism was exhibited by successive presidents of Kenya between 1960 and 2013 in allocating road-building projects. This favouritism disappeared when the country transited into a democracy. Partisan distortions also arise in developed countries such as the United States in the form of ‘pork barrel’ politics, where the allocation of infrastructure projects has been influenced by the partisan bias of national legislators (Levitt and Snyder 1997). Recent literature has devoted increasing attention to the inter-community allocative performance of decentralised governance in a number of developing countries. Personal home bias was demonstrated in an incentive compatible choice experiment by Hoffmann et al. (2017) in the allocation of water infrastructure projects by elected county councillors in rural Kenya between different wards under their jurisdiction. Many other papers show evidence of partisan bias. Azulai (2017) used a triple difference specification to show that rotation of national ministers in Brazil between 2009 and 2016 was associated with corresponding fluctuations in the grants allocated to municipalities controlled by the same political party. Finan and Mazzocco (2017) used a structural approach to estimate a model of allocation of grants across regions within the Brazilian state of Roraima. They found that 25 per cent of these allocations were distorted by electoral incentives relative to social planner’s optimum. In the Indian state of West Bengal, Dey and Sen (2016) showed that following floods in 2013 the dominant party in the state allocated employment programmes to favour those local areas where it won rather than lost by a narrow margin. Shenoy and Zimmerman (2020) used West Bengal data spanning 2011–16 to show that local residents and associated political brokers mobilising votes in the area were rewarded with increased employment benefits when the same party (narrowly) secured control of the local government council. Bardhan et al. (2020) used West Bengal data for an earlier period (2004–11) to show that local government areas redistricted into more competitive state legislature constituencies received larger (or smaller) private benefit programmes if they were controlled by the same (or rival) political party. Studying three South Indian states between 2008 and 2019, Tarquinio (2021) found that drought relief declarations made by state legislators discriminated across local areas based on the electoral competition motives of the party controlling the state legislature. This helped explain why one in three drought-affected areas did not receive any relief, while a third of the relief declarations went to areas without any drought. 3.3 Institutional reforms The distortions described in previous sections raise some obvious questions about possible ways of reforming the institutions of decentralised governance. These include enhanced monitoring and oversight mechanisms, delegating authority to different sets of local agents, and recentralisation initiatives that limit the range of powers devolved to local governments. Enhanced monitoring and verification via e-governance Recent advances in information technology have augmented the capacity of state and central governments to monitor the behaviour of local government officials and control corruption. Largescale e-governance field experiments were carried out in two different Indian states by research teams in collaboration with state government officials in an effort to reduce corruption in NREGA, a programme managed by local governments that employed village residents to build local infrastructure. In Andhra Pradesh, Muralidharan, Niehaus, and Sukhtankar (2016) showed that using biometric identification cards to verify employment beneficiaries nominated by local officials resulted in a 41 per cent reduction in programme leakages associated with ‘ghost’ beneficiaries. Beneficiaries were paid more quickly and reported earnings rose by 24 per cent, while programme costs did not change. Similarly, Banerjee et al. (2019) studied an experiment in the state of Bihar where a manual system of requests for advances from local governments (submitted for verification by higher-tier officials) was replaced by a just-in-time system where payments were automatically triggered by entering the details of beneficiaries in a financial database. The initiative resulted in a reduction of ghost beneficiaries by 5 per cent and in wealth disclosed by upper-tier officials by 14 per cent. At the same time, programme expenditures declined by 24 per cent, while the employment benefits and number of beneficiaries were unaffected. However, after seven months the Bihar experiment was abruptly stopped, in the face of intense pressure from NREGA officials. None the less, the experiment formed the basis of a subsequent nationwide roll-out that realised similar reductions in programme expenditures. It is important to note that the scope of both e-government reforms was limited to reducing corruption (that is, cases where intermediate officials manage to divert benefits into their own private rents, rather than distributing them to local residents). They were not intended to reduce the severity of elite capture or political clientelism, or other forms of misallocation of benefits between local residents. Delegating authority to others One of the main virtues of decentralisation is that it can potentially harness specialised information about local needs and priorities that is possessed by community members or others in close proximity to beneficiaries. Hence an alternative to political decentralisation is to delegate the design and implementation of development projects instead to community leaders or groups. Such community-driven development programmes have been actively sponsored by the World Bank in nearly 90 low- and middle-income countries. 3 These were included in the review of Mansuri and Rao (2013), who noted that such programmes are also frequently subject to elite capture, because community elites can gain more direct control over such programmes. 4 Alternative options include contracting with local members of the community who are hired as ‘agents’, with screening of eligible candidates and performance-based incentives. Examples include government health services contracted with local community agents in Zambia to deliver AIDS prevention services, where the effectiveness of the hired agents varied with their pro-social motivation and the nature of rewards offered (Ashraf, Bandiera, and Jack 2014). Maitra et al. (2020) conducted an experiment in West Bengal where the selection of beneficiaries from an agricultural micro-credit programme was delegated to local community members hired as commission agents (where commissions were linked to loan repayments): here agents were hired either from local trader-lenders or appointed by the local government. While both treatments led to similarly high levels of take-up and loan repayment, only the treatment using trader-lender private sector agents resulted in significant increases in farm income for recipients. These results suggest the need for greater experimentation along these lines. A third option is to delegate implementation to independent non-government organisations (NGOs). While NGOs have played a large and growing role in the implementation of development programmes around the world, there are very few studies comparing their performance to programmes delegated to local governments. A fourth option is for upper-level governments to contract directly with private sector providers, as in the case of grants offered to private schools in Pakistan that competed with traditional government schools (Andrabi et al. 2020). Services delivered by NGOs or private sector providers typically compete with those provided by local governments, resulting in complex interaction effects. They may enhance accountability pressures on government providers due to greater competition or they may result in shrinking government services in ways that end up adversely affecting those citizens who continue to rely on government providers. For instance, in Uganda, Deserranno, Nansambaz, and Qian (2021) found that the entry of NGO health clinics in areas with pre-existing government health workers induced government workers to move to the NGO clinics (which offered higher salaries), resulting in a worsening of infant mortality and overall health care. In areas without pre-existing government health workers, on the other hand, overall health care improved in the absence of any negative spillovers. Limiting discretion over inter-community allocations: formula-based geographic targeting Another type of reform works by restricting the scope for opportunistic manipulation of intercommunity allocations by elected officials, and instead replacing them with formula-based allocations. Three of the papers described above studied inter-community misallocation also examined the likely consequences of this kind of reform. Recall the experimental study by Hoffmann et al. (2017) in which local councillors in Kenya exhibited bias in favour of their own ward in allocating water purification projects. When councillors’ authority to select the inter-ward allocation was removed and replaced by an equal treatment mandate, this home bias was substantially reduced. However, the same intervention raised the councillors’ demand for greater control over the management of the project within each ward, owing possibly to an increase in corruption incentives (that is, via collecting bribes from users within wards). After finding geographic misallocation in funds allocated by Brazilian federal legislators across municipal governments in Roraima state, Finan and Mazzocco (2017) simulated the effect of two counterfactual political reforms intended to reduce partisan bias. They concluded that approval voting would have only small effects on reducing partisan bias. By contrast, introducing single-term limits for legislators was predicted to be more effective in reducing geographical misallocation. However, overall welfare was expected to fall owing to greater diversion of funds to personal interests by single-term incumbents – echoing a concern also raised by Hoffmann et al. (2017) in the Kenyan context. In rural West Bengal, Mookherjee and Nath (2020) used proxy means tests based on household surveys to estimate what improvements in pro-poor targeting would result if grants to village-level governments were based on a formula recommended by the state finance commission rather than the discretion of upper-tier officials. The formula was based on measures of village need – incorporating population, remoteness, literacy, food insecurity, occupational structure, and the population share of scheduled castes and tribes. In the counterfactual policy, benefit allocations within each village would continue to be delegated to respective village governments. The authors found that formula-based grants would actually lower pro-poor targeting slightly. Moreover, altering weights assigned to different criteria used by the Finance Commission could improve pro-poor targeting, but only to a slight degree. This reflects partly the low information content of the criteria used by the Finance Commission to predict the regional distribution of poverty, and partly the political incentives of upper-tier officials to target poorer areas in a clientelistic setting. In addition, the reform would leave unchanged the targeting patterns within villages, where a proportion of benefits end up being allocated to households that are not poor. Removing local discretion: formula-based transfer (private) benefits A more drastic reform would eliminate any scope for local discretion in the allocation of ‘private good’ benefits to households (or individuals), in favour of a formula based on household- or individual-level measures of need, based on the demographic information (for example, gender, age, education, location, family size) available to the government. This would essentially replace existing welfare systems in developing countries with something similar to social security benefits in the US and European welfare states, or the conditional cash transfer programmes that originated in Mexico (Progresa) and Brazil (Bolsa Familia). The centralised system would both determine people’s eligibility for benefits and set up a system to deliver the benefits directly to recipients. The system would not rely any longer on local government officials – their role would thereafter be restricted to providing and managing local public goods. A priori, it is hard to predict what the outcome of such a reform may be. A formula-based programme might be less effective in targeting the poor if (i) local officials are better informed about the incidence of poverty across households and regions than could be captured by the criteria incorporated in the formula; and if (ii) these officials have the incentive to direct benefits to poorer households. But if local officials are either less well informed, or if they have a strategic incentive to target less-poor households (maybe because these are the kinds of households they personally favour, or that would be more responsive with their votes), the reform would improve pro-poor targeting. Some developing countries in Asia have recently been experimenting with reforms of this kind. Haseeb and Vyborny (2021) evaluated a nationwide overhaul of the Benazir Income Support Program (BISP) in Pakistan. Prior to the reform, elected officials identified poor households that would receive the transfers. The reform followed creation of a new administrative database of households and assets with the assistance of the World Bank, which enabled the creation of ‘proxy means tests’ (PMTs) that were then used for targeting instead. The programme effects were identified by differences in the pre- and post-reform changes in the transfers received by households connected to (that is, belonging to the same village and clan as) officials that won elections, compared with other observationally similar households connected to political candidates that had lost elections. This study found that the reform reduced favouritism (that is, the role played by recipients’ connections to elected officials) and improved pro-poor targeting. Moreover, it increased positive perceptions of social protection programmes, both in constituencies that had supported the government and in those that had supported the opposition, with a larger impact on the latter. The costs of implementing the reform, including collection of a nationwide household survey used to create an administrative database, amounted to less than 2 per cent of the total amount paid out in transfers in 2016. A related experiment was carried out in Indonesia by Banerjee et al. (2021). They randomised the roll-out of a formula-based cash voucher scheme replacing an in-kind food assistance programme administered by local government officials. This combined two different dimensions of reform at the same time: cash versus in-kind transfers, and a centralised rather than a decentralised implementation. It is therefore difficult to separate the roles of these two dimensions. However, the combined change resulted in a significant improvement in pro-poor targeting, with a sharper concentration of benefits among poor households (that is, a reduction in errors from including the non-poor). There was a 45 per cent rise in benefits per capita among those who continued to receive benefits. The administrative costs amounted to 4 per cent of the benefits disbursed. These last two experiences suggest that drastic reforms have the potential to improve pro-poor targeting quite significantly. They also indicate that the administrative costs and logistical challenge of implementing such a reform are not large enough to render them infeasible in low-income countries, owing to recent improvements in information technology. 3.4 Rules versus discretion: assessing pros and cons The preceding results suggest the need to consider a rather drastic reform of decentralised governance, where delivery of private welfare benefits would no longer be delegated to elected local officials but instead would use a ‘big data’ approach to create a formula based on PMTs predicting the level of poverty of each household or individual in the country on the basis of administrative surveys. Eligibility for cash or in-kind transfers would be set by the formula, and the programme implemented using a nationwide household or individual-level ID system with biometric identification, combined with direct electronic transfers from central agencies to individuals’ bank accounts or mobile phones. The studies cited point to the difficulties of partial reforms, such as formula-based geographic targeting where local officials continue to retain some discretion (over ‘last-mile delivery’), while the extent of such discretion is curtailed. The benefits of such reforms are that they could enhance targeting to the truly needy, and lower waste and corruption. Additional benefits would include a reduction in the scope of elite capture and clientelism, which could reduce socio-economic inequality within communities, while enhancing political competition and accountability of local government officials. The main responsibility and focus of the latter would be on delivery of local public goods such as roads, irrigation, health, and education, which are likely to have a greater long-run development impact compared to short-term ‘private good’ benefits. What are other potentially important implications of such drastic centralising reforms? The studies reviewed above examine the effect of reforms on targeting in favour of poor households, where poverty is estimated using proxy means tests relying on underlying household attributes in ways that predict consumption expenditures. Such measures can overlook relevant criteria (such as land quality) observed by members of the local community but not captured in household surveys. They also ignore temporary shocks experienced by households, such as illness or crop failure, which lower people’s incomes but are difficult to verify and therefore hard to incorporate into administrative information about household need. Providing security against such shocks is an important objective of government income support programmes. Removing the scope for local discretion by transitioning completely to formula-bound income transfer programmes may then result in a loss of such insurance. How important is this concern? The answer depends on whether decentralised welfare programmes actually do provide insurance against idiosyncratic shocks. The evidence on this issue varies across contexts. In a study of 300 households in central Java, Trachtman, Permana, and Sahadewo (2021) compared community-based assessments of individual households’ needs with those assessed by the households themselves. They found a low correlation (0.16) between community and ownhousehold assessments of weekly (per capita consumption) need, and a higher correlation (0.45) of assessments of asset value. This happens even among families that know each other (for example those who are in close physical and social proximity), suggesting that local communities do not seem to be well-informed about the main temporary shocks experienced by individual households. Community assessments therefore focus on long-term rather than short-term measures of poverty. In Malawi, Basurto, Dupas, and Robinson (2019) evaluated targeting effectiveness (measured by weekly consumption expenditures) achieved in actual allocations of subsidised farm inputs and food aid by tribal chiefs against a counterfactual PMT-based formula. Both the chiefs and the formula failed to identify a substantial fraction of poor people, with chiefs making bigger errors on average. Moreover, chiefs were more likely to target food subsidies to their relatives. On the other hand, compared to the formula-based allocation, the chiefs did provide higher subsidies to people experiencing droughts, floods, cattle death, and crop disease. Hence this study finds evidence that the decentralised allocation provides more insurance than might be achieved by a PMT-based formula. At the same time, chiefs’ allocations were associated with poorer targeting on average, implying that the degree of risk (or inequality) aversion would matter in determining the likely overall welfare impact of such a shift. In similar vein, Dal Bó et al. (2020) found that intermediate supervisors of agricultural extension workers in Paraguay had valuable information about the relative value of distributing scarce cell phones to different extension workers within their jurisdiction, and favoured workers with higher marginal treatment effects when given this discretion in how to allocate them. A scheme that delegates the allocation to supervisors therefore achieves a higher average treatment effect overall, compared to a randomised allocation. However, a more sophisticated centralised allocation using a formula based on the observable characteristics of workers turned out to outperform the delegation mechanism. Administrative challenges While the experiences of Pakistan and Indonesia suggest that the administrative costs of transfer reforms are unlikely to be large, there may be a number of administrative challenges associated with implementation, particularly during the transition to a new system. Policymakers need to create a nationwide identification system; initiate surveys to generate the PMT formula; conduct censuses to collect the asset and demographic data to be used in applying the PMT formula to identify beneficiaries; find ways of integrating this data with the bank accounts or mobile phone accounts to be used for depositing cash transfers; and coordinate with public service institutions involved in delivering any in-kind transfers. In describing the roll-out of the biometric identification smartcards for verifying recipient identity in the Indian state of Andhra Pradesh, Muralidharan, Niehaus, and Sukhtankar (2016, p.2897) stated: After two years of program rollout, the share of Smartcard-enabled payments across both programs in treated subdistricts had reached around 50 percent. This conversion rate over two years compares favourably to the pace of electronic benefit transfer rollout in other contexts. For example, the United States took over 15 years to convert all Social Security payments to electronic transfers, while the Philippines took 5 years to reach about 40 percent coverage in a cash transfer program. In AP [Andhra Pradesh], the inability to reach a 100 percent conversion rate (despite the stated goal of senior policymakers to do so) reflects the nontrivial logistical, administrative, and political challenges of rolling out a complex new payment system. In India, there have been problems in integrating the nationwide identification system, Aadhar, with the public distribution system of food to the poor. The difficulties included denial of benefits to those who lacked the Aadhar ID cards and increased transport costs incurred by recipients, as widely reported in the Indian media and verified in a randomised control trial by Muralidharan, Niehaus, and Sukhtankar (2020). These problems were particularly pronounced for vulnerable groups such as manual workers, widows, and elderly people. The limited inclusion of these vulnerable groups within the financial system has further compounded the problem. Despite the Indian government’s ambitious ‘Jan Dhan’ programme, promoting affordable bank accounts (which was launched in 2014), Pande et al. (2020) estimated that six years later half of all poor women still did not have a Jan Dhan account, and 23 per cent had no form of bank account. A further challenge occurs where individuals’ benefit eligibility is tied to a specific location. Here temporary migrants from rural to urban areas are ineligible for government benefits because their entitlement is tied to their area of origin. In turn, this set-up creates strong disincentives for migration, constraining a significant channel for poverty reduction and growth in developing countries. This problem was highlighted by the acute plight of large numbers of migrants that were trapped in urban areas of India without local support when the Covid-19 shock resulted in sudden loss of employment (Jesline et al. 2021). To create portable benefits that would address this problem, the transfers and social security system in a country would need to be centralised with nationwide identification and eligibility. Broader implications for federalism and democracy Changing to rule-based direct transfer systems would amount to a substantial recentralisation of the public expenditure system in developing countries, lowering the autonomy of state and regional governments in allocating aid. The balance of power between central and regional governments would alter, with possibly significant implications for federalism and democracy. A number of authors have highlighted strategic political economy considerations in decentralisation reforms. Cheema, Khwaja, and Qadir (2006) described how successive waves of decentralisation and recentralisation in Pakistan were related to the conflict between authoritarian rulers at the central level and populist democrats at the regional level. In particular, local governments were created by the former to create a central–local patronage structure that would enable them to bypass regional leaders. In India, since 2014 the BJP-controlled central government has sought to scale back funding for programmes (such as the national rural employment guarantee programme) administered by state and local governments, which appeared to increase the vote share of rival political parties that control state governments, and introduce a set of new welfare programmes more directly identified and controlled by the central government, which raised BJP vote share (Deshpande, Tillin, and Kailash 2019). Martinez-Bravo et al. (2020) provided evidence that in China during the 1980s and 1990s, locally elected village mayors were accompanied by an increase in locally popular policies. Some of these (such as weaker implementation of the one-child policy) were at odds with the goals of central policymakers. After 2000, enhancements of monitoring and bureaucratic capacity of the central government in Beijing enabled them to implement recentralisation reforms that reduced local leaders’ autonomy. These experiences suggest that a transition to formula-based programmes will tend to lower the autonomy and political power of regional governments. If rival political parties control the central and regional governments, this will shift the balance of political power between these parties. Conclusion This chapter has reviewed recent literature pertaining to problems of corruption, elite capture, and political clientelism that limit the scope for decentralised governance to achieve the social welfare goals of cost-effectiveness and targeting the delivery of government benefits to deserving households. A range of reforms intended to curb these problems have varied in their method, scope, and effectiveness. The ones that seemed the most effective involved replacing discretionary authority of locally elected officials over the distribution of targeted transfers of individual/‘private good’ benefits with PMT-based, formula-based transfers. However, the key evidence so far has been limited to only two reform efforts carried out in Indonesia and Pakistan. Studies of these reforms also raised questions regarding the proxy means data that the formulae were based on, their administrative feasibility and some wider concerns for federalism and democracy. Clearly, much remains to be learnt from further experimentation with similar reforms in the future, combined with continued research and discussion about their outcomes.