Central–local relations, accountability, and defensive administration: unraveling the puzzling shrinkage of China’s urban social safety net

Abstract

The calibration between national regulatory oversight and local policy autonomy is a prominent feature in the discussion regarding social assistance institutions in large countries. The complex principal-agent structure embedded within such institutions and the resultant information asymmetry make it difficult for the national principal to monitor the behaviour of subnational agents, resulting in prevalent mis-targeting of welfare benefits and petty corruption. Built on a principal-agent framework, this study seeks to explain the puzzling shrinkage of China’s Minimum Livelihood Guarantee Scheme (Dibao) in recent years. Accountability mechanisms are found to exert major impacts on the scale of welfare programs. Using a unique city-level panel dataset and difference-in-differences (DID) strategy, this quantitative study finds that discipline inspection by upper-level government leads to a significant decrease of Dibao coverage in a city, a link reinforced by the local intensity of China’s anti-corruption campaign. Blame avoidance and the defensive reaction of local agents triggered by draconian enforcement of accountability result in distorted welfare administration on the ground. Building informational capacity presents a useful approach in mitigating the vertical control–autonomy dilemma illustrated in this study.

Introduction

Despite the impressive achievements in global poverty reduction, a significant number of people worldwide are still living below their country’s national poverty line. The coronavirus disease 2019 (COVID-19) pandemic and the resultant economic catastrophe over the past 4 years have pushed many people back into poverty, raising the demand for social protection. New labour market risks, created by the technological revolution, automation replacement and the thriving platform economy, have intensified such welfare demands (Busemeyer & Sahm 2022; Au-Yeung et al. 2025). The past decade has witnessed a divergent trajectory of social protection programs in terms of coverage and generosity. Whilst some countries have expanded the scale of such programs, notable retrenchment is seen in other countries. Such variation reinvigorates the debate in social policy theories: what explains welfare expansion and retrenchment (Niedzwiecki and Pribble 2025)?

China offers a fascinating case in this scholarly debate due to the puzzling shrinking in recent years of its flagship social assistance program, the Minimum Livelihood Guarantee System (MLG), or Dibao. Instituted as the last-resort social safety net, Dibao is the world’s largest cash transfer program in terms of beneficiary population (Kakwani et al. 2019; Lammer 2023). However, recent years have shown notable evidence of a reverse-U shape in its nationwide coverage, precipitously dropping from 75.8 million beneficiaries in 2011 to 40.3 million in 2022 ¹ .

Importantly, this drastic decline occurred amidst the Chinese government’s repeated call for increase in benefits and coverage. What explains this remarkable discrepancy between the mandate from Beijing and what happens on the ground? Is this significant reduction of welfare mainly attributable to China’s massive poverty alleviation campaign that claimed to have eliminated absolute poverty in the country by 2020? Or were the local governments who administer the program tactically defying the central government mandate? This paper seeks to answer these research questions through a quantitative investigation.

Much of the old wisdom tends to view the state as monolithic entity in social policy making and welfare provision, often neglecting the complex principal-agent structure embedded within the administrative system. In large countries with a multilevel hierarchy, dynamic central–local interactions profoundly shape the provision of social welfare (Huang 2015; Huang & Kim 2020). It is well recognised that local agents may not necessarily share the same social policy pursuits as the central principal even in authoritarian systems that maintain rigid political control (Gong & Wu 2012; Li 2006). This incongruence is often complicated by fiscal arrangements between national and subnational authorities (Ratigan 2017; Zhu 2016). Often driven by electoral motivation, local states in liberal democratic societies are held accountable by the local community for their welfare delivery, whilst national-level social legislations also exert pressure for subnational compliance. In authoritarian systems where national policy guidelines are supposed to have a greater power in shaping local welfare scale, some recent studies have revealed a variety of ‘gaming’ behaviours of the local state, arguably driven by political or fiscal self-interest (Lin & Dale Tussing, 2017; Zeng 2020; Guo, He & Wang 2022).

Whilst the classic principal-agent theory still holds significant explanatory power in elucidating subnational social policy choices, this present study enriches the theoretical discussion with a new perspective: accountability mechanisms. Our empirical study in China underscores the critical importance of disciplinary enforcement exercised by the central state in shaping local welfare provision. Specifically, our quantitative evidence suggests that discipline inspections exercised by provincial government as a response to the central government mandate has led to a significant decrease in the coverage of urban Dibao. In addition, the intensity of China’s anti-corruption campaign has substantially reinforced this link. This study argues that local governments and frontline administrators as rational agents of central government – the political principal – adopt blame-avoidance strategies to minimise risk in the face of sweeping accountability pressure. The rising conservatism towards new Dibao applications as well as the draconian dismissal of existing beneficiaries have resulted in potentially large errors in both exclusion and inclusion. A key message arising from this study is that a healthy social protection institution must be built on solid accountability, but campaign-style enforcement may trigger defensive behaviour by frontline administrators that undermine the social legitimacy of the institution itself. Given the vast asymmetric information between the central principal and local agents, agile enforcement of accountability should hinge on strengthened informational capacity whilst taking street-level bureaucratic capacity and informal rules into account.

The remainder of this paper is organised as follows. Section 2 delineates the empirical context, highlighting the puzzling patterns of China’s urban Dibao program in recent years. Section 3 establishes the theoretical framework, leading to the formulation of research hypotheses in Section 4. Research design and empirical results are reported in Section 5 and Section 6, respectively. Section 7 discusses the results against theories and policy practice. Section 8 concludes the paper with policy implications and limitations.

Empirical context: Dibao and its puzzling shrinkage

First launched in the City of Shanghai as a local pilot, Dibao was officially instituted in 1999 by the Chinese government as a key social assistance program for urban China. The program was scaled up nationwide at the turn of the new millennium in response to the vast restructuring of China’s state-owned enterprises that were causing hefty unemployment and financial hardship for numerous urban households. The initial motivation of Dibao was to secure the smooth progress of urban economic reform and minimise social instability (Solinger & Hu 2012; Gao, Yang & Li 2015). Mounting social needs, coupled with the strong fiscal capacity of the state in the Hu Jintao leadership, led to a rapid expansion of urban Dibao that subsequently inspired the launch of rural Dibao in 2007. Covering 75.8 million residents in both urban and rural areas in 2011, Dibao is the world’s largest cash-based social assistance system (Li & Walker 2018).

Issued by the State Council in 1999, the first regulatory framework of Dibao stipulated that ‘urban residents with non-agricultural household registration status, if the average income of their family members is below the minimum living standard of local urban residents, are entitled to material assistance from the local government for their basic life’ (State Council 1999). Providing cash transfers to eligible residents, Dibao is designed to maintain their basic living standard defined as ‘necessary costs of food, clothes, and housing, given reasonable consideration to water and power and fuel bills, and educational costs for children’ (State Council 2014).

Although the policy framework is set by the Ministry of Civil Affairs at the centre, the actual operation of Dibao follows a highly decentralised approach, with the structure supposed to balance national policy rigidity and local flexibility. Local governments are responsible for setting eligibility thresholds and payment standards, screening applications and the disbursement of benefits. Governments above county level are required to set a Dibao threshold computed in accordance with the local minimum living standard and the financial capacity of the local government. To be eligible, an applicant must demonstrate that their household income falls below the locally determined assistance line. A means-test is administered by local government offices, considering the income, savings and assets of an applicant. A set of administrative procedures were set in place to determine eligibility, including screening, verification and neighbourhood peer review (Huo & Lin 2019). Several empirical studies have noted positive outcomes of Dibao in alleviating urban poverty (Wu & Ramesh 2014; Gao, Yang & Li 2015).

Initially, local governments were required to finance Dibao but the central policymakers soon found that many cash-strapped local governments were struggling to run the program (Solinger 2014). The central authorities responded by offering special fiscal transfers to provinces with financial difficulties, and this fiscal arrangement led to a dramatic expansion of Dibao coverage until 2009 (Ngok 2013; Xu & Carraro 2017). Central fiscal transfers accounted for more than 80% of expenditure on Dibao by the end of 2018. ²

The puzzle inspiring this study is the dramatic contrast between the Chinese government’s continuous policy efforts in expanding the Dibao program and its astonishing shrinkage since 2010. The average Dibao threshold in the cities saw a 20.8% annual increase between 2007 and 2022, from CNY 2,188.8 to CNY 9,027.6. The income replacement rate (monthly Dibao payment/average individual income) of urban Dibao doubled during the same period, whilst that of rural Dibao saw a two-fold increase. ³ Clearly, the Chinese government has shown a strong political commitment to strengthening the social assistance system. However, in reality there was a significant continuous shrinkage of Dibao (see Figure 1 below).

Figure 1. Total population and threshold of Urban Dibao (2007–2022). Source: China Civil Affairs Statistical Yearbook (various years).

Was there a big reduction in poverty in China during this period? Yes and no. Commentators often attribute the remarkable downscaling of Dibao to China’s landmark poverty reduction campaign that claims to have eliminated absolute poverty in the country in 2020. ⁴ However, the campaign primarily took place in rural China, whilst the vast urban areas were not subjected to similar nationwide policy interventions. What we saw was a striking shrinkage of urban Dibao over thirteen consecutive years, now covering less than 1% of the urban population. Given the marked slowdown of China’s economic growth and rising urban unemployment in the aftermath of the COVID-19 pandemic, such a rapid decline of coverage may not be news worth celebrating. The Chinese government has responded since 2020 with several policy campaigns, striving to expand the coverage, but unfortunately these have yielded rather disappointing outcomes. By the end of 2022, a rebound in coverage had been observed only in Fujian, Ningxia and Qinghai, with the rest of China still witnessing sustained shrinkage. ⁵ Note, this drop persisted despite central government’s substantive relaxation of the eligibility threshold during the pandemic that strived to bring more vulnerable urban residents into social protection (Lu et al. 2020).

Was this shrinkage due to fiscal austerity, given the Chinese government’s struggle with the economic downturn and drop in revenue? No. Quite the contrary, the reality is that the central government continuously increased fiscal expenditure to the Dibao program, from CNY 84.5 billion in 2009 to CNY 250 billion in 2022. ⁶ Ironically, the massive unspent Dibao fund – amounting to CNY 1.29 billion in 2021 – has been a major policy concern since 2015. Some provincial governments, notably Guangdong, Hainan and Shanxi, were even reprimanded by central authorities for their significant amount of unspent Dibao budget. ⁷

The facts summarised above depict a perplexing situation: there is evident mounting demand for social assistance and the central state is willing to spend (and financially capable of doing that) but a sizable budget remains unspent on the ground. This puzzle can hardly be explained by fiscal disincentive or intentional welfare retrenchment alone. Instead, this paper seeks an answer from within the bureaucracy, one that is deeply embedded into China’s central–local relations, as to how accountability pressure alters the incentives of local agents in welfare provision.

Theoretical framework and institutional context

The classical principal-agent theory and the Weberian theory of bureaucracy present useful frameworks to explain the challenges and complexities that arise in the relationship between a principal (typically a central government) and an agent (typically a local government), particularly in the delegation of authority and the pursuit of shared objectives (Lane 2005; Miller 2005). One key aspect of this theoretical framework is the misalignment of goals and incentives between the principal and the agent. The central government often prioritises national objectives, whereas subnational governments tend to focus on local interests and preferences. This divergence in goals can lead to conflict and tension in implementing national policies at local level (Tanner & Green 2007; Chung 2000; Li 2006). Local governments may also face pressure from their constituents, which may differ from the priorities set by the central government (Lin & Dale Tussing, 2017; Guo, He & Wang 2022; Huang 2015). Another key dimension of this theoretical framework is the recognition of information asymmetry between the principal and the agent. Monitoring the actions of local agents may be theoretically possible, but gathering complete information is often prohibitively expensive, particularly in large developing countries. This information asymmetry can result in grave challenges in coordinating policies, implementing programs and achieving desired outcomes, such as expanding social programs (Liu, Xin & Yi 2024; Zhou & Lian 2020; Liu et al. 2009).

The principal-agent theory also highlights the concept of agency costs – the costs incurred by the principal to ensure that the agent acts in its best interest. The principal may need to invest resources in monitoring, auditing and controlling the actions of local agents to minimise the risks of opportunistic behaviour or non-compliance (Miller 2005; Weingast 1984). These agency costs can include the establishment of reporting mechanisms, conducting performance evaluations and enforcing accountability measures. To mitigate agency problems, the central government employs various control mechanisms, such as policy guidelines, targeted fiscal transfers, oversight structures and disciplinary action. Accountability and responsiveness are crucial elements in the principal-agent relationship (Lane 2005).

In theory, by employing appropriate control mechanisms and fostering accountability, the central–local relationship can be strengthened, leading to more effective governance and service delivery. Yet the exercise of accountability rules is often a formidable mission as it hinges on the possession of good information regarding the behaviour of subnational agents. Thus, accountability enforcement can be costly and time-consuming, requiring significant resources and high administrative capacity (Miller 2005). This can pose a challenge for central principals with limited capacity and information infrastructure, particularly in developing countries.

The case of China presents a vivid illustration of the challenges summarised above. Despite the unitary nature of the polity and strong central control, waves of administrative decentralisation in the 1980s and 1990s have created de facto fiscal federalism in the country’s central–local structure under which the local states gain considerable negotiation power with the central principal (Shieh 2000). Sluggishness, goal substitution, collusion and outright defiance were frequently observed in the local implementation of national policies since the reform era. The central state responded to these challenges with extensive use of fiscal transfers and funded mandates, but non-compliance appeared to persist. Gong and Wu (2012) argued that whilst heightened central fiscal capacity was expected to increase the likelihood of local compliance, it in fact does not, because the discretional power remained in the hands of local authorities.

Further aggravating the accountability challenges are the prevalence of informal societal rules and the weak rule of law. Opportunism in local agents tends to proliferate when the divergence of incentives widens between central and local actors. In the case of Dibao administration, a variety of collusive behaviours have been well documented in the literature, including misreporting data, collusion with other local agents, manipulating information and exploiting loopholes in implementation (Li & Walker 2021; Zhou & Lian 2020; Tang, Wang & Yi 2023). It must be noted that many of such activities remained frequently observed in the past 10 years despite the sweeping political recentralisation in China, thus highlighting the robustness of local opportunism. When explaining the local non-compliance of central mandate, Gong and Wu (2012) contend that whilst central mandates are derived from paramount political authority, the lack of powerful monitoring and accountability measures gives rise to persistent local non-compliance. Yet this present study will demonstrate the pendulum nature of central–local relations in the management of China’s social safety net: draconian accountability enforcement and disciplinary monitoring may not necessarily lead to outcome desired by the central state.

Observations and hypotheses

Deviant administration and challenges in monitoring

Despite its wide coverage, Dibao has long been criticised for its low targeting accuracy (Du & Park 2007; Gao et al. 2009; Li & Walker 2018; Golan, Sicular & Umapathi, 2017). Kakwani et al. (2019) estimate that rural Dibao excluded 87% of the poor whilst including 82% of the non-poor, and that the percentage of resources leaked out to the non-poor is even larger than the leakage of recipients. According to the estimate of Song et al. (2022), the exclusion error of urban Dibao falls between 38.45% and 66.28%, whilst its inclusion error is between 54.59% and 69.17%. A variety of perverse behaviours has been documented in abundant anecdotal materials as well as scholarly research. The existence of such widespread mis-targeting gravely challenges not only the cost-efficiency of the fiscal resources used, but also the social legitimacy of the policy as a whole (Li & Walker 2018). Reflecting significant misuse of government funds, the mis-targeting of Dibao is caused by local governments’ self-interest and the practical challenges of regulatory oversight.

Local governments, particularly those in middle- and low-income regions, have a strong motivation to include more residents into Dibao and thus expand its scale, given the generous fiscal transfer from the centre (Li & Walker 2021). Social assistance entitlements are often co-opted by local governments and community leaders as a ‘perk’ to pacify dissidents and maintain social stability, as well as to build political support, foster social harmony and encourage community participation (Solinger & Hu 2012; Li & Walker 2018). In addition, many studies have found ‘street-level bureaucrats’ taking Dibao allowances for their family members and relatives (Han & Gao 2019; Li & Walker 2018; Golan, Sicular & Umapathi 2017). Endemic deviations such as giving Dibao allowances on the basis of ‘human feelings’ (renqing bao), social connections (guanxi bao) and the need for maintaining social stability (weiwen bao), as well as through false claims (pian bao) and mistakes (cuo bao), have become prevalent in practice (Lammer 2023; Golan, Sicular & Umapathi 2017). Most of such cases of ‘welfare capture’ involve manipulating applicants’ information (Han & Gao 2019). Furthermore, information asymmetry also exists between applicants and the local Dibao administrators, as reporting of household income is easily subject to fraud and hidden income from informal employment (Xu & Carraro 2017).

Monitoring mechanisms have been put in place, but the significant information asymmetry makes it hard for central government to carry them out. When ad hoc spot checks come from above, close kinship makes it easy for street-level bureaucrats and local residents to collude (Zeng 2020). The multiple layers of principal-agent relations, compounded by challenges in effective monitoring, fuelled the rapid expansion of Dibao in the first decade of the new century (Li & Walker 2018). Central government was not blind to these irregularities and responded in two ways. First, considerable efforts were made in building up information infrastructure. Once launched, a big data platform can perform means-tests and data authentication with high accuracy (Zhu, Xiao & Wu 2024). However, building up a nationwide big data platform is a technically formidable task requiring a comprehensive integration of massive household-level data related to individuals’ economic activities. It is particularly challenging to interconnect the database of commercial banks as well as other financial institutions.

Accountability measures and consequences

In the absence of reliable data, the central government had to respond to rampant mis-targeting through large-scale inspection campaigns. Since 2013, the Communist Party of China (CPC) Central Discipline Inspection Committee, the National Supervisory Commission and the Ministry of Civil Affairs launched several waves of inspections. Accountability was particularly strengthened against the backdrop of China’s landmark ‘targeted poverty alleviation’ (TPA) campaign championed by top leadership. In fact, the launch of the ambitious TPA reflected Beijing’s frustration with the existing approach of social assistance, including Dibao (Li & Walker 2021). Suspicion of significant misuse of the poverty alleviation fund led central government to tighten up accountability measures in 2013 (Zeng 2020). In October of that year, the Ministry of Civil Affairs prescribed a series of monitoring methods and established a register of Dibao recipients who are close relatives of Dibao administrators or community/village leaders. Targeted disciplinary inspections followed. These inspections into Dibao coincided with the CPC’s massive anti-corruption campaign, creating a highly tense political environment for local governments and numerous civil servants. Local officials responsible for Dibao administration were facing unprecedented pressure from potential disciplinary penalties and even legal consequences. From 2018 to 2022, there were close to half a million corruption and disciplinary misconduct cases related to livelihood affairs including Dibao all over the country, with 456,000 officials subjected to party or administrative penalties. ⁸

In a typical inspection campaign, sophisticated protocols of financial auditing, household visits, cross-checks and anonymous reporting are set in place, followed by high-profile propaganda and rigid accountability rules. Both the discipline inspection authorities and the Ministry of Civil Affairs stressed zero tolerance to various forms of maladministration of Dibao. ⁹ Under such mounting political pressure, a natural reaction from local officials was to rectify widespread mis-targeting by dismissing many ‘undeserving’ recipients. For example, in one of the campaigns in 2014, 64.29 million Dibao cases were reviewed, with 4.16 million recipients losing their entitlement (officially called ‘exit’ from Dibao), including 257,000 cases related to erroneous inclusion. This campaign alone brought 12,631 cases into disciplinary investigation. ¹⁰ In another such campaign in 2019, 1.85 million recipients nationwide were dismissed in 3 months. ¹¹ In Inner Mongolia, an inspection campaign in 2015 led to the punishment of 200 local officials, and Dibao coverage dropped by 150,000 individuals the following year. ¹²

Such high-powered accountability pressure has yielded unintended consequences. Realistic concerns of blame avoidance have driven many local officials to switch to a passive stance towards Dibao administration. Both qualification screening and approval of applications have been substantially tightened up at grassroots level, resulting in the exclusion of many in need. Our fieldwork in five provinces (Ningxia, Hainan, Guangdong, Heilongjiang and Shanxi) found marked hesitation on the part of civil affairs officials in approving new Dibao applications, and an equally worrisome trend is the ‘forceful exit’ of many deserving recipients. Box 1 below exhibits some of the examples we came across in the fieldwork that indicate an unjustifiable exit from Dibao. According to some interviewees, ¹³ the limited discretionary power in their hands is now associated with unlimited responsibility. The comment of one interviewee in a coastal China province is very representative of the opinions we collected from multiple regions:

This type of inspection is meant to be strict. Uniform rules are applied everywhere. In practice, we must drop all existing recipients so long as there is one family member having a permanent job in government, even though they are truly poor…The central government is now exclusively focussed on mis-targeting. We [local civil servants] are held accountable for any targeting error. The consequence [of being caught] is that you must claim back all the Dibao money from the household, and you have to pay from your own pocket if you fail to do so. Given such a harsh circumstance, we would rather keep a tight hand in processing new applications. You effectively reduce the liability on your shoulders when you approve fewer applications. When accountability inquiry kicks in, you are not even given an opportunity to explain. ¹⁴

Box 1. Real cases of Dibao application/exit decisions

Case #1.

Through a household visit, the applicant was found to own a car. Eligibility was immediately rescinded. In fact, the impoverished applicant had bought this used car for CNY 8,000 and was trying to feed the family through running a ride-hailing service.

Case #2.

Data verification discovered that the household applicant has bank savings of CNY 14,690, which is CNY 1,010 higher than the upper limit of savings stipulated by the local regulations for the means test. Eligibility was consequently rescinded. The fact was that there was only one old couple in this poor family. Extremely frugal and worried about sickness, the impoverished couple struggled to save money for healthcare.

Case #3.

Case review found that the household was covered by the old-age pension program. Dibao entitlement was immediately revoked. In fact, they live in a poor city that used to be dominated by state-owned heavy industries. As a result of massive bankruptcy of state-owned enterprises, a big population of urban workers were laid off. Applicants in this household were unemployed and had been struggling with their livelihood for several years.

Case #4.

An impoverished old man was living alone. Case review found that his son has a stable job that should have disqualified the man from Dibao. The fact is that the son had never fulfilled his familial responsibilities and the father had not received financial support from him. The Inspection Team hence required an immediate revocation of his Dibao entitlement. Unless the elderly father declares termination of paternity at the local court, the Team insisted on his ineligibility.

Case #5.

Case review found that the applicant owns a company. Dibao entitlement was immediately revoked. In fact, he was merely the nominal head of the company, which is owned by someone else. This non-performing small company does not have any actual operation or revenue.

Case #6.

A poor household had been a Dibao recipient for several years. The sick father was living with children, left behind by the mother who left over 10 years ago and never came back. Dibao regulation requires the household combined income to be below the local threshold. However, due to the absence of necessary divorce record, the actual income of the ex-wife could not be substantiated. The Inspection Team thus required an immediate revocation of the household’s Dibao entitlement whilst the officials-in-charge were subject to administrative punishment.

Source: Authors’ fieldwork in Heilongjiang Province and Hainan Province.

Hypotheses

The observations above delineate the peculiar interaction between accountability enforcement and welfare scale. When lenient enforcement of accountability is coupled with self-interest, local governments and street-level bureaucrats have considerable incentives to expand Dibao coverage, regardless of pervasive mis-targeting, forming a robust community of collusion (Li & Walker 2021). The limited information capacity of central government in monitoring the program operation leaves it little choice but to resort to managed campaigns – the old-fashioned tool in the arsenal of the Chinese party-state (Liu et al. 2015; Zhou & Lian 2020; Zeng 2020). Rigid enforcement of accountability – campaign-style discipline inspection in its Chinese form – powerfully alters the incentive structure of local governments, who are the administrators of the program. As the previous collusion has been largely broken down with government officials at all levels, with each held accountable both prospectively and retrospectively, a high degree of fear was instilled through all levels of subnational government (Li & Walker 2021); discretion dwindled. Switching to a defensive stance of Dibao administration, blame-avoiding local governments and their officials therefore now tend to take a much more stringent approach to program entry and to opportunistically accelerate program exit. To test this speculation, we formulated the following hypothesis:

H₁: Ceteris paribus, discipline inspection initiated by the central government leads to significant scaling-down of the Dibao program in a city.

In the meantime, as we stressed above, multiple rounds of Dibao inspection co-occurred with China’s sweeping anti-corruption campaign. Various studies have found a remarkable decrease of local government innovation and a rise in conservatism of local officials due to the high political risk created by the campaign (Chen & Göbel 2016; Heffer & Schubert 2023). Although Dibao-related misconduct accounts for a very small fraction of convicted corruption cases, we posit that the anti-corruption campaign still forms the broad political environment for numerous frontline Dibao administrators. We expect that an increase in its intensity would strengthen the natural defensive mentality that, in turn, drives local governments towards conservatism, at the expense of potentially large exclusion errors. Hence, we hypothesise:

H₂: An increase in the intensity of China’s anti-corruption campaign in a city reinforces the downscaling of its Dibao program.

Research design

Data source

The empirical evidence for this study comes primarily from a large prefectural-level panel dataset on the Dibao population from 2009 to 2019. The gigantic Dibao program has been a popular subject of social policy research in the past two decades, but a key methodological shortcoming is that the existing quantitative literature is predominantly based on national or provincial-level data. This data constraint has potentially masked the micro dynamics at the sub-provincial level, given the significant responsibility of city and county governments in administering Dibao. Our study rectifies this shortcoming by using a unique dataset of city-level statistics obtained from the Ministry of Civil Affairs.

The full dataset covers the period between 2008 and 2021, but the observations of 2008 were excluded due to many missing values and data definition issues. The observations of 2020 and 2021 were also dropped because the unusual change in the Dibao population was inevitably affected by the COVID-19 pandemic. As a result, the final panel dataset covers the Dibao population in prefectural cities between 2009 and 2019. There are 333 prefectural administrative divisions in China, comprising 293 cities, 7 equivalent districts and 33 ethnic-minority autonomous prefectures. The last two categories are typically represented by those located in remote ethnic-minority areas of small populations and low economic status, and including them in the statistical analysis could introduce unnecessary bias to the sample. Therefore, we focussed exclusively on prefectural cities that account for the majority of prefectural administrative entities in China. As a small number of cities (N = 13) underwent an administrative merger or other jurisdictional changes that complicate the identification of the Dibao population, we also decided to drop these observations from our sample. In the end, the working sample was comprised of 274 prefectural cities over 11 years of continuous observations (cross-sections).

Variables

Denoted as ‘Dibao-scale’, the dependent variable counts the number of urban Dibao recipients in a city of a given year. Accountability pressure is represented by two independent variables, namely, ‘discipline inspection’ and ‘anti-corruption intensity’. The former is coded in binary terms where value 1 means that the provincial authorities ¹⁵ initiated an urban Dibao inspection in its jurisdictions of a given year, and value 0 otherwise. Data for this independent variable were collected from websites of local governments and official newspapers. The variable of ‘anti-corruption intensity’ is measured by the total number of duty-related crimes (e.g. corruption, bribery, negligence, abuse of power, etc.) per 10,000 public personnel in a province of a given year. As this variable indicates the annual volume of judiciary verdicts, it well captures the temporal intensity of anti-corruption pressure in a province. Specifically, the number of duty-related crimes was based on official statistics reported by the local procuratorate of each province. The number of public personnel in a province was collected from the China Labor Statistical Yearbook. It is noteworthy that 2019 saw many missing values on duty-related crimes, and we therefore used the data between 2009 and 2018 for this variable.

Several city-level variables were controlled for in statistical analysis, including: (1) eligibility threshold of urban Dibao in a city, (2) per capita disposable income of urban residents, (3) size of urban population, (4) local gross domestic product (GDP), (5) fiscal income of prefectural government, (6) number of unemployed people in a city, (7) average wage of current urban employees, (8) fiscal expenditure on urban Dibao and (9) unspent budget on urban Dibao. Data for these control variables were collected from official statistical yearbooks.

Variable definition and descriptive statistics are presented in Table 1 and Table 2, respectively.

Table 1. Variable definitions

Note: Data for 2009–2017 is based on city-level data. Ministry of Civil Affairs has ceased releasing city-level data on Dibao threshold since 2018, and therefore we used provincial average for observations of 2018 and 2019.

Table 2. Descriptive statistics

Empirical strategy

Urban China has not been subjected to significant policy interventions in poverty alleviation in the past years, whilst a certain percentage of the cities experienced targeted discipline inspection that is posited to exert impact on local Dibao scale. Thus, the research setting well resembles a quasi-experiment in which Dibao inspection is the ‘treatment’, whilst prefectural city is the unit of analysis. The method of difference-in-differences (DID) was employed to test H₁. On the one hand, Dibao inspection campaigns may lead to considerable behavioural changes in local government discretion that in turn affect the scale of Dibao. On the other hand, mis-targeting and the Dibao population may differ significantly between cities that had discipline inspections and those that did not. Hence, the DID strategy can account for the effects of other simultaneous policy interventions, as well as the a priori differences between inspected and uninspected cities. In this way, Dibao inspections can be seen as a quasi-experiment and the DID method is able to help estimate the net effect of such ‘treatment’ in the Dibao scale. Specifically, the estimation strategy is presented by the formula below:

Dibao-scale _it = α+ θinspection _it +λ Z_it + η _i + μ _t + ε _it

where Dibao-scale _it refers to the total number of Dibao recipients of city i in year t whilst X is a dummy variable indicating whether city i experienced Dibao inspection in year t. η _i and μ _t represent city-level fixed effects and year fixed effects, respectively. ε it denotes the random error term that may affect discipline inspection in a city. θ, the DID estimator, captures the impact of Dibao inspection on the urban Dibao population. Z _it γ loads the control variables that affect Dibao population of city i in year t.

We further included the interaction term for anti-corruption intensity as a moderator to test H₂. Here, we paid due caution to possible multicollinearity between ‘discipline inspection’ and ‘anti-corruption intensity’. Following the protocol suggested by Aiken and West, we performed a mean-centring procedure for both variables before the interaction term was generated and included into estimation models.

Empirical results

Baseline results

The DID method is conditioned on the parallel trend assumption. Therefore, we must prove that before accountability pressure was intensified, the changing trend of the Dibao population ran in parallel across cities. We adopted the event study method proposed by Jacobson et al. (1993) to perform the parallel trend test. As Figure 2 shows, there was no statistically significant difference between cities that later had discipline inspections and those that did not, in terms of size of Dibao population. Hence, the parallel trend assumption is met.

Figure 2. Result of parallel trend test.

Table 3 reports the baseline statistical results. Model 1 reveals a significant negative effect of discipline inspection in the Dibao population of prefectural cities. Specifically, cities subjected to inspections were associated with an average decrease of 5,126 recipients against those not inspected. Model 2 includes the independent variable, moderator and control variables into regression analysis, and the statistical patterns remain robust. With other factors considered, discipline inspection still led to a significant average reduction of 5,388 recipients. Thus, H₁ was endorsed. Model 3 examines the moderating effect of anti-corruption intensity between discipline inspection and the urban Dibao scale. Aside from the two independent variables, the interaction term was also included into regression analysis. The statistical significance of the interaction term suggests that anti-corruption intensity strengthened the negative impact of discipline inspection on the local Dibao scale. H₂ was hence supported.

Table 3. Baseline regression results

Note: Robust standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

The operational mechanism of Dibao inspection was given due attention in our statistical analysis. Such inspections typically take place in the middle of a year, intended to leave sufficient time for local governments to take corrective actions and to lift the targeting rate to a decent standard. Because it may take time for local governments and Dibao administrators to alter their behavioural patterns following the change in their incentive environment, the actual impact on Dibao scale may see a time-lag effect. To account for this possibility, we lagged the key independent variable by 1, 2, 3 and 4 years, respectively, and examined the statistical patterns again. As presented in Table 4, the shrinking effect of discipline inspection lasted for up to 2 years and vanished from the third year. Looking at the coefficients alone, the magnitude of the effect appears to be even stronger in the second year than the first year. The message underlying this result is two-fold. First, disciplinary pressure does not entrench into the frontline overnight, but it takes time – arguably through social learning and peer effect – for local governments and frontline administrators to make behavioural adjustment on the basis of the new accountability environment. Second, the ‘magical’ disappearance of the shrinking effect from the third year appears to imply that forceful exit of Dibao may have reached the limit. In other words, the shrinking effect of disciplinary pressure is temporary in nature.

Table 4. Baseline regression results (time lag effect of discipline inspection)

Note: robust standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1

Robustness checks and additional identification tests

The baseline regression results indicate that inspected cities were more likely to see shrinkage in their urban Dibao. We performed robustness checks to corroborate the main results. Results of the checks are reported in Table 5 below. In model 1, the number of Dibao recipients is replaced by the coverage rate ¹⁶ of urban Dibao in percentage terms. This ratio represents a relative measure of the Dibao scale useful for cross-city comparison. In model 2, discipline inspection on rural Dibao replaces the key independent variable. Both the urban and rural programs are administered through the same organisational structure, thus we believe that accountability pressure created on the rural program will inevitably spill over to the management of the urban program. Due to the small coverage of Dibao, the regression coefficient becomes too small. Hence, we algorithmised the control variables for ease of interpretation. The results are consistent with those of the baseline models.

Table 5. Results of robustness checks

Note: robust standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

To address the concern that the baseline DID results may not establish a causal link between discipline inspection and the local Dibao scale, we performed two additional tests. The difference in the Dibao scale between the treatment group (inspected cities) and the control group (uninspected cities) could be caused by natural changes over time. To account for this possibility, we conducted a placebo test by randomly assigning fictitious event times into the sample. Specifically, we advanced the time of discipline inspection by 2 years, 3 years and 4 years, thus constructing a false policy timeline. Table 6 presents that the estimated coefficients of the placebo did not pass the significance test at the 0.1 level. Hence, there is no systematic difference in the time trend between the two groups of cities.

Table 6. Placebo tests using pseudo-event year

Note: Robust standard errors in parentheses; *** p < 0.01, ** p < 0.05, * p < 0.1.

Next, we conducted another placebo test by randomly assigning treatment and control cities. The rationale behind this test was: if the inspected cities indeed reduced their Dibao scale, this effect should only exist in real treatment cities. In the placebo test, we randomly selected the same number of cities as the false treatment group and false control group. The impact of the placebo can be obtained through the estimated coefficients. We then repeated the above process 500 times to obtain 500 regression coefficients and their corresponding p-values. By plotting the kernel density distribution and p-value of these 500 estimated coefficients, the regression coefficients fall above 0 and follow a normal distribution, with most regression results being statistically insignificant (Figure 3). This test suggests that our main results are unlikely to be driven by chance.

Figure 3. In-space placebo test.

Discussion

This study has sought to explain the puzzling shrinkage of the Minimum Livelihood Guarantee Scheme, the world’s largest cash-based social assistance program, through the lens of central–local relations and accountability mechanisms. Using a principal-agent framework, the study theorises local governments as rational agents of the central principal, and their choices with regard to Dibao administration are jointly determined by the incentive structure and accountability environment. Welfare capture of local administrators and community elites creates the micro-foundation for widespread leakages (Han & Gao 2019; Li & Walker 2018). Local governments also take a utilitarian stance towards Dibao administration given the generosity of central fiscal transfers. It is clear that local government as an institution and frontline administrators and community leaders as self-interested individuals formed a collusion to expand the coverage of Dibao until 2010 under the environment of loose central control. The asymmetrical information between multiple layers of the colossal hierarchy makes it extremely difficult for the central principal to exercise effective monitoring. Weak enforcement of accountability hence offered ample space for nepotism and other forms of maladministration at the grassroots level.

The unprecedented strengthening of accountability since 2013 was in part a reaction to the central principal’s inability to monitor the micro-operation of Dibao. Campaign-style enforcement kicked in, unexpectedly triggering other forms of strategic reaction at the local level. Discipline inspection was frequently launched, and the pressure penetrated to street-level bureaucrats against the backdrop of an overwhelming anti-corruption campaign in the country. Active dismissal of existing recipients and passive admission of new recipients combined to cause the dramatic decrease of urban Dibao coverage in the past years. The explanation above has been supported by our quantitative data analysis using the DID method. The results still held after three additional tests were performed.

This study sheds fresh light to the social policy literature. The calibration between regulatory oversight and local policy autonomy is well known in the discussion regarding social assistance institutions, especially in the developing world (Ditch et al. 2018; Li & Walker 2021). Whilst the importance of necessary local discretion is appreciated in running such programs in large countries, petty corruption, clientelism, funding leakages and embezzlement are widely observed (Alatas et al. 2012; Muralidharan et al. 2016; Han & Gao 2019). Multi-layer hierarchy is particularly prone to collusive behaviour amongst local agents, complicating effective monitoring by the principal. Our study from China illustrates that swinging the pendulum towards centralised accountability may backfire. Draconian disciplinary enforcement gives rise to defensive administration of social assistance that is not aligned with the principal’s goals, highlighting the persistence of the ‘vertical control-autonomy dilemma’ (Ahlers & Schubert 2015; Liu, Xin & Yi 2024). This accountability perspective allows us to further theorise the behavioural patterns of subnational agents in welfare provision.

Concluding remarks

On the basis of this study, we conceptualise defensive administration as a behavioural response by public officials – particularly street-level bureaucrats – that emphasises the minimisation of personal and institutional blame within the context of service delivery. This approach is characterised by a focus on adhering to formal regulations and maintaining procedural correctness, often at the expense of responsiveness to the unique circumstances of the public they serve. The resultant administrative practices reflect a self-protective mentality that prioritises the safeguarding of individual and organisational reputations over effective governance.

This study draws crucial implications for social assistance policies in China and beyond. To mitigate the vertical control-autonomy dilemma, it is of critical importance to reduce the deep information asymmetry between the multiple layers of the principal-agent structure: national policymakers, subnational implementing agencies, street-level bureaucrats and social assistance beneficiaries. One promising approach in reducing asymmetrical information is the development of information networks through which information can flow swiftly between stakeholders. In recent years, there has been prominent progress in big data platforms enabling information-sharing across government departments to detect possible deviation in administering social assistance. Such tools have shown notable effectiveness in reducing mis-targeting and opportunistic behaviour (Zhu, Xiao & Wu 2024; Zeng 2020).

Whilst it is our key argument that big data-empowered monitoring presents a useful approach in reconciling the innate tension between national regulatory uniformity and the flexibility of local discretion, we do recognise the limits of this line of solutions. Entrenched local bureaucratic interest compounded by the vast informal ‘rules of the game’ on the ground may work to practically subvert technological solutions. There is anecdotal evidence showing how street-level bureaucrats respond to digital monitoring through new forms of bureaucratic formalism. ¹⁷ It is hence a caveat of this study that the power of information infrastructure should not be overestimated, particularly in a context characterised by significant disparities in administrative capacity and civil service professionalism.

This study is certainly not without limitations. For obvious reasons, the actual number of corruption cases in a locality is unknown to researchers. It is plausible that the strength of anti-corruption campaign was not uniformly applied throughout the country and the targeting of inspections followed certain patterns. However, these possibilities cannot be substantiated without access to really high-level political ‘low-down’, which was unfortunately unavailable to us. As a result, our measurement of the independent variables may be subject to certain bias. Moreover, despite the abundant qualitative data collected from our multi-province fieldwork, we are only able to include very limited qualitative insights due to rigid word limit. It is our plan to systematically present such analysis in a companion paper. We humbly acknowledge these limitations and will try to address them in future research.