2.1. Two approaches to weak enforcement

Weak policy enforcement is a common problem in the developing world. Existing explanations tend to follow two approaches: low state capacity and selective enforcement, or forbearance.

The capacity-based approach attributes weak policy performance to limited state capacity. A weak state may undersupply resources to enforcement agencies, poorly select or monitor its agents, or lack embeddedness with social groups that foster compliance (Evans, Reference Evans1997; Fearon, Reference Fearon, Przeworski, Manin and Stokes1999; Tsai, Reference Tsai2007; Rothstein and Varraich, Reference Rothstein and Varraich2017; Dasgupta and Kapur, Reference Dasgupta and Kapur2020). Applying this logic to the Chinese context, the capacity-based explanations usually focus on the administrative capacity of the social insurance agency (Tang and Feng, Reference Tang and Feng2019), institutional fragmentation and the principal-agent problem (Zhuang and Ngok, Reference Zhuang and Ngok2014), and lack of compliance from firms and employees (Mao et al., Reference Mao, Zhang and Zhao2013; Rickne, Reference Rickne2013). For example, Tang and Feng (Reference Tang and Feng2019) find that by switching the responsibility of collecting social insurance payments to the tax agency, which generally has a higher administrative capacity than the social insurance agency, social insurance extraction can be better enforced.

However, this approach assumes that stronger capacity will automatically lead to better enforcement. What it overlooks are bureaucratic agency and incentives. Even high-capacity states may choose not to enforce certain policies. For example, East Asian developmental states—a type of state known for its high-quality bureaucrats and strong capacity—tend to have poor records of enforcing labor-protecting policies (Caraway, Reference Caraway2009; Song, Reference Song2012), a problem that seems to go beyond the lack of capacity.

The forbearance approach remedies some inadequacies of this literature by shifting focus from what the state can do to what it chooses to do. Scholars argue that states sometimes intentionally withhold enforcement, despite having the capacity to do so. For example, in Latin America, politicians may tolerate street vending and squatting as an informal form of redistribution that helps win votes (Holland, Reference Holland2016). Here, enforcement becomes a strategic choice, influenced by competing priorities and political incentives (Zacka, Reference Zacka2017; Dewey et al., Reference Dewey, Woll and Ronconi2021).

Moreover, the forbearance approach highlights that weak enforcement is not a failure per se, but can serve broader policy goals (Dewey et al., Reference Dewey, Woll and Ronconi2021; Feierherd, Reference Feierherd2022; Lopez-Cariboni, Reference Lopez-Cariboni2024). For example, authorities in Latin America relaxed labor policies in response to trade liberalization and increased competitive pressure (Ronconi, Reference Ronconi2012). Similarly, labor standards were lifted or selectively enforced in Italy to address broader social challenges, such as unemployment and immigration (Ceccagno, Reference Ceccagno2017).

However, much of this literature is democracy-focused, relying on electoral accountability as the incentive for forbearance (Smulovitz, Reference Smulovitz2015; Feierherd, Reference Feierherd2022). In authoritarian regimes, where leaders face fewer pressures from voters and seek to maintain tight control, forbearance is more puzzling. Autocrats are not typically seen as responsive to societal demands (De et al., Reference De, Bueno, Smith, Siverson and Morrow2005). Moreover, forbearance requires the state not to sanction noncompliers, which seems detrimental to authoritarian rule. Weak enforcement discourages quasi-voluntary compliance by suggesting that violators will not be caught, which may add to the cost of social control significantly (Levi, Reference Levi1989). The incentives of forbearance in authoritarian states therefore need to be more closely examined.

A few recent studies have shed light on forbearance in authoritarian states. Lopez-Cariboni (Reference Lopez-Cariboni2024) shows that autocratic leaders offer less forbearance regarding electricity theft during economic shocks than their democratic counterparts, and argues that it is because of fewer electoral incentives and longer time horizons. Yet there can be other forms of political incentives for autocratic leaders to forbear. Shen (Reference Shen2022), for instance, examines regulatory forbearance toward China’s polluting industries. She argues that while politicians in democracies are reelection-maximizing, their Chinese counterparts are promotion-maximizing and use forbearance to maximize their chance of getting promoted.

A related literature on authoritarian forbearance has delved into labor protection. Scholars have found that most East Asian authoritarian states—such as Singapore, Malaysia, and Vietnam—tend to have more protective laws than the rest of the world while achieving fewer actual protections (Caraway, Reference Caraway2009, Reference Caraway2010). The gap between de jure and de facto labor protection in the region has been attributed to weak enforcement of the laws. However, the underlying reasons for this weak enforcement remain ambiguous, and there’s a lack of direct examination into the roots of authoritarian forbearance.

We build on this literature by theorizing that in China, forbearance of social insurance policies stems from bureaucratic incentives rather than electoral accountability. Specifically, local officials have career incentives to promote economic growth and appease firms, often at the expense of social insurance enforcement, to improve their promotion prospects. Ironically, while forbearance in developing democracies may function as informal welfare for the poor, in China, it undermines formal protection and yields regressive consequences on Chinese workers.

2.2. Weak enforcement of China’s social insurance policies

We develop a theory of forbearance under autocracy through an application to China’s social insurance policies. China is the largest authoritarian country in the world and its working population plays a key role in global production. Enforcement of social insurance policies in China is highly important (Huang, Reference Huang2013, Reference Huang2014; Gallagher et al., Reference Gallagher, Giles, Park and Wang2015; Yang and Shen, Reference Yang and Shen2021), as weak enforcement has eroded the country’s safety net and fueled social unrest.Footnote ⁴

China’s modern social insurance system was introduced in the 1990s, replacing a socialist-era scheme that primarily covered SOE workers while excluding those in the growing non-state sector (Li et al., Reference Li, Feng and Gizelis2007; Zhou and Zhang, Reference Zhou and Zhang2017). Economic reforms during this period also led to massive SOE layoffs, sparking widespread protests and threatening regime stability (Chen, Reference Chen2000; Cai, Reference Cai2002; Hurst and O’Brien, Reference Hurst and O’Brien2002; Zhu, Reference Zhu2025). In response to these challenges, the central government launched reforms that established five key insurances, including pensions, healthcare, unemployment, work injury, and maternity. They were extended to cover all urban employees, including those in the private sector (Zhang et al., Reference Zhang, Zhang and Bingwen2019). A major institutional change was the introduction of a cost-sharing model in which employers, employees, and the state all contribute, with employers being the primary contributors.

To enforce premium collection, the state enacted a new labor law in 1998 and the Interim Regulation on the Collection and Payment of Social Insurance Premiums (1999–2019) in 1999.Footnote ⁵ These measures transferred enforcement responsibilities from labor unions to local governments, typically at the city level.Footnote ⁶ Local authorities were granted discretion over whether the social insurance bureau or the tax bureau would serve as the collection agency.

Despite central mandates, local enforcement was notably weak. By 2004, the coverage rates for pensions, medical, unemployment, and work injury insurances among urban workers were only 36.6%, 27.8%, 23.7%, and 15.3%, respectively (Wu, Reference Wu2023). High contribution costs deterred compliance: in 2008, employers in Shanghai paid an additional 39.5% of a worker’s salary for social insurance (Wu, Reference Wu2023). Firms often evaded these costs with worker consent, offering slightly higher wages instead of paying long-term benefits. Moreover, the social insurance bureau was the primary agency for collecting premiums at that time, which may have faced capacity constraints. Unlike the tax bureau, the social insurance bureau lacks access to extensive firm information, allowing room for firms to misreport data on employee numbers and salaries and evade social insurance payments.

In an effort to improve enforcement, the state implemented the Labor Contract Law in 2008, mandating firms to incorporate social insurance obligations into labor contracts.Footnote ⁷ Following the Labor Contract Law, the state also introduced the Social Insurance Law in 2010, outlining the obligations of both employers and employees regarding social insurance contributions. Moreover, the state gradually transferred the responsibility of collecting social insurance premiums from the social insurance bureau to the tax bureau, thereby enhancing enforcement capacity (Tang and Feng, Reference Tang and Feng2019).

These measures, while reduced the number of firms completely evading social insurance contributions, had limited effectiveness in ensuring that firms pay the full amount of premiums. Instead, firms can still partially evade social insurance obligations. For example, a local government may establish standards for wage base on which social insurance premiums should be collected, usually with the minimum wage base set at 60% of the local average income. Many firms then choose to underreport employees’ wages and only pay the minimum required amount. Additionally, some firms may create new accounts or benefits for employees, avoiding depositing their entire salary into designated salary accounts.

2.3. Bureaucratic forbearance and how it works

Persistent weak enforcement of social insurance policies cannot be explained by capacity constraints alone. Instead, evidence suggests that local governments often intentionally withhold enforcement. This bureaucratic forbearance is evident in three ways. First, enforcement is chronically weak: Nationwide surveys show that a large share of firms do not comply with mandated contribution levels. For example, the China Enterprise Social Insurance White Papers reported that 61% of firms in 2015 did not pay premiums based on employees’ actual wages. This figure further rose to 75% in 2016 and 76% in 2017.

Second, enforcement is uneven across jurisdictions and firms. With significant autonomy in policy implementation, some localities have set employer contribution rates below the national standard (e.g., lowering the 20% pension rate) to attract investment and maintain economic competitiveness. Some city governments have also made explicit commitments to business owners, allowing them to take gradual steps to enroll in social insurance programs and only pay the minimum amount required.Footnote ⁸ Moreover, enforcement is selective across firms. Studies find that domestic private firms have faced especially lax enforcement compared to SOEs, allowing them to “avoid as much as possible” (Wu, Reference Wu2023).

Third, enforcement is inconsistent and reversible over time. Qualitative evidence shows no consistent enforcement mechanism at the local level. Local officials primarily “resort[ed] to the politics of persuasion” to encourage firms’ compliance (Duckett, Reference Duckett2001, p. 298). Specifically, they often “had to resort to negotiations and a combination of repeated pleas” to convince some firms to join the program (Duckett, Reference Duckett2001, p. 228). In addition, punishments often lack credibility: Firms found to have made no contributions were simply asked to pay the arrears, and there was also “no effective strategy for targeting of defaulters in subsequent audits” (Nyland et al., Reference Nyland, Smyth and Zhu2006, p. 201).

Moreover, these enforcement efforts were easily reversible, as they depended heavily on bureaucratic initiative and could be withdrawn through reduced effort or suspended audits, even after formal enforcement mechanisms were introduced. Such reversals are often framed by policymakers and state media as necessary to reduce business burdens and support economic growth, especially during economic shocks.Footnote ⁹ This was particularly evident during the COVID-19 pandemic, when growth pressures intensified. In Wuhan, for example, local officials waived all employer pension contributions for 2020.Footnote ¹⁰ In Shanxi, authorities extended deadlines, allowed deferred payments, and instructed agencies to treat firms as compliant during the grace period.Footnote ¹¹ Our data similarly show that some firms contribute in one year but not the next. For instance, among firms that paid unemployment insurance in a given year, approximately 18% ceased payment in the subsequent year. The reversal rate was around 9.5% for medical insurance and pensions. These figures suggest that enforcement is not only weak and uneven but also subject to ad hoc reversals.

The administrative structure of the social insurance system further facilitates this forbearance. To enroll in the social insurance system, firms must register with the local collection agency, open a designated bank account, and submit information on employee headcount and salaries. Once approved, they are expected to pay premiums monthly and update records regularly.Footnote ¹² In practice, however, the system relies heavily on self-reporting, and these records can be easily manipulated. Firms may under-report the number of workers or their wages, or simply avoid updating records when hiring or granting raises, especially given that audits are infrequent and discretionary (Nyland et al., Reference Nyland, Smyth and Zhu2006; Wu, Reference Wu2023). Workers may not be aware of reduced or missing contributions, as employer payments typically go into pooled public accounts rather than individual ones.Footnote ¹³ In some cases, workers and firms even collude: Workers may willingly forgo benefits in exchange for higher take-home pay, especially if they are uncertain about receiving future payouts (Gallagher et al., Reference Gallagher, Giles, Park and Wang2015; Yang, Reference Yang2021).

Can weak, uneven, and reversible enforcement be driven by firm-side noncompliance alone? We argue that even if firms are the immediate decision-makers in suspending or lowering contributions, this behavior is likely responsive to local officials’ behavior or expectations about enforcement. Firms respond to bureaucratic signals (Lin and Milhaupt, Reference Lin and Milhaupt2021; Lin, Reference Lin2025). In regions where officials signal leniency, firms anticipate low risk of audits or penalties and adjust their behavior accordingly. In this sense, noncompliance is not solely a firm-driven act of evasion but a joint outcome shaped by local enforcement behavior.

2.4. Political root of bureaucratic forbearance

When would a local government choose not to (strictly) enforce social insurance policies? A key consideration is local officials’ career incentives. Unlike elected officials, Chinese local officials are selected and promoted by their upper-level principals. Promotion is highly competitive, as officials at the same level need to compete for a limited number of positions at the upper level (Lü and Landry, Reference Lü and Landry2014). The Chinese cadre evaluation system heavily emphasizes economic performance, particularly GDP growth rate, as a primary criterion for advancement (Wang, Reference Wang2015; Landry et al., Reference Landry, Xiaobo and Duan2018). Regardless of whether GDP growth directly results in promotion (Maskin et al., Reference Maskin, Qian and Chenggang2000; Li and Zhou, Reference Li and Zhou2005; Xu, Reference Xu2011; Shih et al., Reference Shih, Adolph and Liu2012; Jia et al., Reference Jia, Kudamatsu and Seim2015; Jiang, Reference Jiang2018; Bo, Reference Bo2019), the system creates strong shared perceptions that growth is essential, prompting officials to prioritize it above other goals.

Under this system, officials are highly sensitive to GDP growth pressure, which can be understood in absolute and relative terms. Absolute pressure reflects pressure against oneself and is shaped by tenure. Although local officials nominally serve 5-year terms, actual tenure is often uncertain. As a result, officials tend to intensify growth-promotion efforts the longer they remain in office, making “years in office” a useful proxy for absolute pressure (Keng et al., Reference Keng, Pang and Zhong2016). While the tenure-based approach captures absolute pressure, it overlooks that promotion also depends on relative performance. As Lü and Landry (Reference Lü and Landry2014) note, with a fixed number of leadership positions, officials must not only meet performance benchmarks but outperform their peers, i.e., other city leaders in the same province.

We argue that the relative performance approach more accurately reflects promotion pressure in China. As Zhou (Reference Zhou2004) points out, political advancement functions as a zero-sum tournament, in which one official’s success indicates the other competitors’ failure. Therefore, what matters is not absolute but relative performance. Since city leaders typically compete within the same province, falling below the provincial average in GDP growth intensifies the pressure to stimulate the local economy.

Forbearance on social insurance policies has been widely perceived as an informal pro-business policy, which local officials believe will benefit the local economy and thus their careers.Footnote ¹⁴ In this case, forbearance has distinct advantages compared to other pro-business policies. Unlike tax breaks (Chen and Zhang, Reference Chen and Zhang2021) or subsidies (Hou and Li, Reference Hou and Siyao2023) that draw from the current government budget, the costs associated with social insurance forbearance are borne by a large number of employees, with its detrimental effects only possible to manifest in the future. Forbearance can also be flexible and revokable (Holland, Reference Holland2017), enabling government officials to adjust based on their own interests.

Will forbearance backfire on officials’ promotion? Local officials often face the challenge of balancing two objectives: achieving economic growth and maintaining social stability. One might ask whether under-enforcement of social insurance could lead to unrest such as large-scale worker protests, and therefore pose risks to local officials’ career advancement.

We argue that while local officials take protests seriously, those triggered by social insurance grievances are relatively rare and manageable.Footnote ¹⁵ China’s authoritarian system maintains a highly effective apparatus of control and repression, which makes collective action difficult. Local officials often rely on tools such as surveillance, intimidation, and the detention of activists to preempt or suppress unrest (King et al., Reference King, Pan and Roberts2013; Xu, Reference Xu2021), without needing to change long-term enforcement behavior. By contrast, growth-related incentives are more immediate, measurable, and institutionalized within cadre evaluations, and therefore often outweigh the potential risk of protest.Footnote ¹⁶

Based on the theory developed above, we derive the following hypothesis:

2.5. Selectivity in bureaucratic forbearance across firm type

Local officials not only determine “when” to grant social insurance forbearance but also “to whom,” based on their career incentives. We anticipate that the impact of GDP growth pressure on inducing forbearance varies among three types of firms: SOEs, domestic private firms, and foreign firms. To begin with, GDP growth pressure is unlikely to induce forbearance among SOEs. SOEs operate under direct government oversight and encounter less severe market pressures. Moreover, they are mandated to prioritize “social responsibility” with stability at its core. Therefore, SOEs may not be the direct target of bureaucratic forbearance on premium collection.

In contrast, career incentives have prompted local officials to grant forbearance to domestic private firms and foreign firms. Domestic private firms employ the majority of low-skilled labor in China and face intense market competition (see our statistics in the next section). They are integral to the local economy, prompting officials to extend them forbearance on social insurance. Moreover, foreign firms are important engines of local economic growth, and their investment and innovation are also crucial indicators in the cadre evaluation system (Chen, Reference Chen2018). Hence, promotion-seeking officials are likely to alleviate their burden of labor costs through forbearance.

Therefore, we derive the following hypothesis:

3.1. Data and methods

We leverage the CIES from 2001 to 2007. Collected by China’s National Bureau of Statistics (NBS), CIES is one of the most comprehensive micro-level databases on industrial firms in China. It includes all industrial firms with annual sales exceeding 5 million yuan (710 thousand USD), representing the most granular level at which firm statistics in China are directly collected rather than estimated.Footnote ¹⁷ The sum of firm-level industrial indicators from CIES corresponds to 95%–98% of the aggregate amount reported by the National Statistical Yearbook. In addition, the CIES relies on firms’ direct and mandatory reporting to China’s central statistical authority. That is, all above-scale industrial firms are required to submit detailed financial and operational data through standardized forms (Nie et al., Reference Nie, Jiang and Yang2012), thereby avoiding potential manipulation by local officials (Wallace, Reference Wallace2016).

Most importantly for our purpose, the CIES reports contributions to social insurance programs at the firm level. This is one of the few firm-level census datasets that include such information in an authoritarian context, making it a valuable resource. Consistent with existing qualitative evidence, firms were pretty open about their nonparticipation in social insurance programs across cities and years. This is likely because noncompliance with social insurance contributions has been widespread and often normalized—a common perception in China is that “the law does not punish the masses (fa bu ze zhong).” Furthermore, as mentioned earlier, the enforcement is mostly lenient. Specifically, noncompliance rarely results in meaningful penalties: even firms identified through audits were typically only required to make retroactive payments, with little evidence of systematic follow-up or effective targeting of defaulters in subsequent audits (Nyland et al., Reference Nyland, Smyth and Zhu2006). Finally, the NBS is a statistical authority rather than a regulatory one. It does not have the mandate or power to directly penalize firms based on their reports, and it is in fact strictly prohibited to share sensitive firm-level information for purposes other than “statistics” (i.e., data collection, organization, and analysis).Footnote ¹⁸ These data are also not publicly disclosed, which further reduces the perceived risk of self-reporting noncompliance.

We measure our outcome of interest in two ways: a dummy variable indicating whether a firm participated in social insurance programs in a given year, and a continuous variable calculating the rate of contribution for participating firms (contribution at time t divided by total wage at t − 1).Footnote ¹⁹ Given the available data in CIES, we create these variables for both unemployment insurance (2001–2007) and medical insurance and pensions (2004–2007) as our outcome of interest. It is also worth noting that our measurement strategy follows the same logic as in studies such as Tang and Feng (Reference Tang and Feng2019). Even if firms systematically overreported their social insurance contributions, our measures would represent an upper bound of compliance, which further underscores the prevalence of forbearance. Moreover, we would generally expect overreporting to attenuate the estimated effects of interest in our firm-level analysis. In other words, the potential presence of overreporting likely leads us to present conservative estimates.

Figure 2 shows the time trend of firms’ participation rate in social insurance schemes. Panel (a) demonstrates that the average participation rate in unemployment insurance has been fluctuating around 40% across the country. The average participation rate in medical insurance and pensions in the last three years was higher but still about 60%. Additionally, we find that almost 80% of firms contributed less than 10% of total wage payments to medical insurance and pensions combined, which is significantly lower than the 20% threshold required for pensions alone in most cities. Given that the nation- and province-wide stipulations have been in place for multiple years, it is surprising to see the persistence of such low contribution rates. Panel (b) disaggregates the trend by ownership types. Consistent with our expectations, SOEs took the lead in social insurance participation, while domestic private firms performed the worst. The implications for the labor force are tremendous, considering that domestic private firms accounted for over 80% of the firm population and 70% of employment in the world’s most populous economy.

Figure 2. Firms’ participation rates in social insurance across years and ownership types. Panel (a) shows firms’ average participation rates in unemployment insurance, medical insurance, and pensions across all available years. Panel (b) demonstrates firms’ average participation rates in unemployment insurance across years and ownership types.

For our key treatment variable, we follow existing literature by measuring interjurisdictional competition as the lagged difference in GDP growth rates between provincial averages and individual cities. We label this variable as GDP growth pressure whose larger positive values indicate greater pressure from peers within the same province.Footnote ²⁰ As discussed in our theory section, this measure can better capture the “relative” performance of local leaders, which may play a bigger role in political promotion. That being said, we also include well-known measures of promotional pressures such as absolute GDP growth rates, turnover years, and the number of years in office. The next section offers further discussions about alternative measures of interjurisdictional competition, and our results remain robust.

Finally, we collect a battery of control variables from multiple sources. We first include firm-level characteristics that are important for firms’ participation rates in social insurance, including total output, employment, wages, and profits. We then merge the dataset with the China City Statistical Yearbook that contains key information about government deficits, business taxes, aggregate FDI, etc. Last but not least, we add individual attributes of Chinese mayors from the Chinese Government Official Data, including education, ethnicity, and gender. The variable about political connections with provincial superiors come from Jiang (Reference Jiang2018). Appendix A provides the summary statistics of our variables.

3.2. Interjurisdictional competition and bureaucratic forbearance

To examine our key hypothesis, we first conduct firm-level analysis by regressing firms’ participation dummy and contribution rate on interjurisdictional competition across cities in the same province. We subset our sample for non-provincial capitals, because (1) mayors of provincial capitals might be of higher ranks and (2) provincial capitals enjoyed natural advantages in GDP growth competitions and were hard to compare with other cities. As mentioned earlier, we control for a series of covariates at firm, city, and mayor levels, as well as firm and year fixed effects.Footnote ²¹ All continuous variables are log- or inverse hyperbolic sine-transformed when skewed. Overall, our model specification leverages fine-grained variation at the firm level, allowing us to estimate how a given firm changes its social insurance payment due to the changes in citywide GDP growth pressure over years. The causal identification relies on the assumption that with our rich set of control variables, there are no other time-varying confounders. Unless otherwise noted, all of the standard errors in our statistical analysis are clustered at the city level, the level where GDP growth pressure took effect.

To account for the capacity-based explanation—the major alternative to forbearance, we adopt several approaches. First, we include in our regression a common measure for government capacity, extraction of business taxes (% of GDP). If local governments were able to extract more taxes from each firm, they should, at least in theory, also possess a stronger capacity to enforce premium collection. Therefore, we label this variable as Capacity. Second, as some provinces have delegated the authority of premium collection from social insurance bureaus to tax bureaus, we include a dummy (Tax Agency) to represent such transition. Since tax bureaus have been perceived to be more capable of collecting payments from firms, their takeover would indicate a major boost for bureaucratic capacity (Tang and Feng, Reference Tang and Feng2019). Finally, we leverage other objective measures of capacity such as the number of doctors and hospitals in a given city-year in our analysis of medical insurance and pensions.

The statistical results in Table 1 confirm our hypothesis that interjurisdictional competition significantly lowered firms’ participation in and contribution rate to these insurances. The effect size is also notable: Based on the binary measure in columns (1) and (3), being left behind by 4 percentage points in GDP growth rate (the average pressure level) led to at least a 2.4 percentage point decrease in a firm’s participation in the given city-year. This would translate to about an 11 million yuan decrease in a typical city-year in China during this period. Added together, this accumulated to significant deficits in social insurance funds as we observe today. Moreover, columns (2) and (4) show that economic competition also lowered firms’ contribution rates. Firms tended to shrink social insurance coverage for their employees when promotion incentives prompted government officials to forbear.

Table 1. Interjurisdictional competition and social insurance (firm level)

Note: *p < 0.1; **p < 0.05; ***p < 0.01.

Our results also lead to interesting implications. We observe that the negative effect of “absolute” GDP growth might suggest that harder economic conditions led to wider coverage of these insurances, consistent with our conventional wisdom. However, if “relative” GDP growth pressure was also present, then bureaucratic forbearance could in fact offset the former effect. This effect turns out to be more significant and substantial. Even though these two variables are moderately negatively correlated (see Appendix A.3), we disentangle two distinctly different effects, where “relative” performance often dominates “absolute” performance in reducing firms’ labor costs induced by social insurance policies.

In contrast, most of the capacity variables do not have a statistically significant and positive effect on firms’ social insurance participation or contribution rate. The extraction of business taxes appears insignificant in explaining premium collection based on the estimated coefficients on Capacity. We also find that the transition to tax authority actually led to more rampant forbearance for firms’ participation, though they might extract more premiums for medical insurance and pensions from existing participants. Finally, in Appendix B.2, we also find no significant effect of city-level medical resources on firm-level participation in medical insurance and pensions.

Our firm-level analysis exploits within-firm variation in social insurance contributions over time and alleviates many concerns about firm-level confounders. Furthermore, we conduct a series of additional analyses in Appendix B. We find consistent empirical support for bureaucratic forbearance when using the full sample without winsorization, allowing more flexible and non-linear functional forms, accounting for dynamics (leads and lags) in two-way fixed effects models, controlling for firms’ political connections via subsidies and state capital, considering party secretary characteristics and labor strikes, and excluding new firms, firms with extreme values, and those with changed ownership or location. We also find consistent evidence when considering firms with multiple branches as well as suggestive evidence for the implied effects of social insurance contribution on GDP growth.

Firm-level data on social insurance contributions is only available through 2007, due to changes in CIES questions. The timing aligns with the implementation of the 2008 Labor Contract Law, which aimed to strengthen labor protections, formalize employment relations, and enhance enforcement of employer obligations, including social insurance contributions. To examine whether similar patterns of bureaucratic forbearance persisted in the wake of the Labor Law and under new leadership after 2013, we conduct additional analysis using individual-level data from 2012 to 2015 in the next section. Although these datasets are not directly linked, they allow us to triangulate evidence from multiple sources and time periods to assess the robustness of our theoretical argument.

3.3. Individual-level evidence for bureaucratic forbearance

We now supplement our firm-level analysis with individual-level surveys. Specifically, we leverage the CFPS, a nationally representative survey launched by Peking University. Introducing this survey data brings unique advantages to our analysis. On the one hand, our CFPS data extend from 2012 to 2015, enabling us to test our hypotheses even after the 2008 Labor Law and the 2013 leadership turnover. On the other hand, individual-level surveys help us avoid potential misreporting by firms, offering a convincing robustness check.

We “replicate” our firm-level analysis with the individual survey data. We first limit our analysis to all firm employees across all occupation categories. We then create an outcome variable indicating whether a firm employee received any of the three types of social insurance in a given year: unemployment insurance, medical insurance, or pensions. Once again, although these individuals are entitled to social insurance coverage through their employers, over three quarters of them report not participating.

Compared with firm-level analysis, our main specifications control for individual fixed effects in addition to city and year fixed effects, enabling us to account for mostly constant demographic variables across these years that might explain social insurance coverage, including education, gender, party membership, and hukou (household registration) status. Individual fixed effects also help address concerns that unobserved, stable characteristics—such as personal preferences, work ethics, or other cultural traits—might confound the relationship between city-level economic development and individual-level outcomes. Substantially, we are “fixing” one person and estimate whether she is more or less likely to obtain social insurance when she lives in the city-year experiencing higher GDP growth pressures. Most city- and mayor-level variables remain consistent with those used in the previous firm-level analysis.Footnote ²² We expect that GDP growth pressure at the city level would decrease the probability of firm employees securing their social insurance.

Table 2 presents consistent evidence for our main hypothesis. When city-level economic competition became more intense, firm employees in the given city-year were less likely to secure these social insurances, even after controlling for individual fixed effects and other city and official covariates. We also find that the effect of absolute GDP growth is less significant and the effect of capacity is not significant. Hence, the impact of bureaucratic forbearance extends beyond the firm level to directly affect individuals employed by these firms.

Table 2. Interjurisdictional competition and social insurance (individual level)

Note: *p < 0.1; **p < 0.05; ***p < 0.01.

3.4. Selectivity of bureaucratic forbearance

To explore heterogeneous effects across different subgroups of firms, we differentiate firms based on ownership types. We do so by investigating firms’ ownership types at the time of their initial registration. Although there are different ways to categorize ownership types in China, we choose to use the registration type as it is most directly related to a firm’s social insurance status. We then conduct a subsample analysis using the same model specification as in our firm-level analysis.

Figure 3 presents the estimation results. Consistent with our hypothesis, the effects of GDP growth pressure on social insurance forbearance are significant and substantial for private firms, whether domestic or foreign, whereas such effects are statistically indistinguishable from zero for SOEs. Local officials were likely to forbear the former groups of firms as they relied more heavily on affordable labor. These heterogeneous effects have unintended distributional consequences. First, it widened the gap between SOE employees and workers in the private sector in terms of their social insurance protection. Second, given the demographics of workers in different firms, employees in private firms (especially domestic ones) tend to be more vulnerable. The number of low-skilled workers in domestic private firms is three times higher than that in SOEs,Footnote ²³ making them more susceptible to job market uncertainties. Therefore, these effects highlight how China’s new social insurance scheme might further stratify the labor market and even exacerbate inequality, the very problem it was designed to address (Huang, Reference Huang2013).

Figure 3. Effect of GDP growth pressure on social insurance participation across different ownership types (subsample analysis). GDP growth pressure is negatively associated with social insurance participation across firms of different ownership types, especially for private and foreign firms.

4.1. Different measures of interjurisdiction competition

As shown earlier, we find that the “relative” performance of local officials better captures promotional pressures and thus the underlying incentives to forbear firms’ social insurance payments. While we have leveraged GDP growth rate differentials between provincial averages and individual cities, it is possible that local officials were more susceptible to pressures from neighboring competitors. For example, Luo and Qin (Reference Luo and Qin2021) find that the probability of promotion is negatively associated with the economic performance of neighboring jurisdictions. Therefore, for each city, we recalculate our measure of GDP.

Growth pressure by taking the difference between its GDP growth rate and that of its closest city, or the average difference from its two closest cities. In Appendix C.1, we uncover similar results showing that economic competition has led to more pervasive forbearance in social insurance policies.

4.2. Different administrative levels

While the overall policy guideline has designated premium collection to city-level governments, official commentaries suggest that county-level officials also had substantial discretionary power in certain localities.Footnote ²⁴ To check whether bureaucratic forbearance mostly took place among city officials, we “replicate” our analysis at different administrative levels. Specifically, we aggregate our firm-level variables to the city level and estimate the same specification with the same set of covariates in Appendix C.2.1. In addition, we calculate GDP growth pressures among counties and re-estimate our firm-level models in Appendix C.2.2. County panel covariates are from China County Statistical Yearbook. The results confirm that the city-level GDP growth pressure significantly reduced social insurance coverage at the conventional level, whereas the county-level GDP growth pressure did not. This provides consistent evidence that city-level governments were primarily responsible for collecting social insurance premiums and had the discretion to forbear firms’ payments during competitive years.

4.3. Does labor matter?

While local officials might forbear firms’ social insurance payments in the hope of better economic performance, such forbearance was exercised at the cost of workers. If workers were able to organize collective action for their right and welfare based on the law, it could severely threaten the career prospects for local bureaucrats. Hence, local officials should at least balance their approach to firms and workers. To test this implication, we collect reported labor strike data at the city level from StrikeMapFootnote ²⁵ and merge them with our individual-level survey data based on all the years available. We estimate the same model but with lagged labor strikes as an additional explanatory variable.

As presented in Appendix C.3, we find that labor strikes do not have a substantial effect on firms’ social insurance participation or contribution. Nor do they alter our substantial findings much. This is consistent with the observation that the large amount of labor in China has had limited bargaining power vis-a-vis government officials or firms (Gallagher, Reference Gallagher2004). Of course, it is not to say that labor strikes were inconsequential to local officials. It was more likely that local officials managed to control the impact of labor protests at their early stages while selectively co-opting workers through other means of compensation than social insurances. All in all, labor seemed to have limited power in negotiating their social insurances that were critical during economically difficult times.