1. Introduction
How do social relationships between government officials shape the flow of information within bureaucratic hierarchies? Across political science and economics, scholars have long emphasized that lower-level bureaucrats often hold preferences that diverge from those of their principals (Stiglitz, Reference Stiglitz1989; Pollack, Reference Pollack1997; Gailmard and Patty, Reference Gailmard and Patty2012). Classic principal–agent theories highlight information asymmetries—gaps in what principals know about agents’ actions—as a central source of inefficiency and bureaucratic drift. These asymmetries allow agents to pursue their own interests, create obstacles to oversight, and hinder effective governance. As a result, much of the existing literature has focused on diagnosing the negative consequences of information gaps and designing mechanisms to curb them, such as monitoring, changing incentives, and institutional reforms.
Yet, despite this rich body of work, we know far less about the underlying nature of information asymmetries themselves and the actual communication practices within bureaucracies. What kinds of information do lower-level agents choose to share—or withhold—from their superiors? What factors could impact communication between government officials?
This article addresses these questions by examining how social relations between officials influence internal communication. Specifically, we argue that patron–client ties embedded within formal hierarchies are critical in explaining the information flow between government bureaucracies. When lower-level officials are clients of higher-level patrons, they have incentives to engage in informal information campaigns designed to portray themselves as diligent and competent. In contrast to the conventional view that information asymmetries uniformly obstruct oversight, we contend that the presence of patron–client relationships can actually increase information flows from agents to principals. In these cases, social ties and bureaucratic authority overlap in ways that encourage communication rather than impede it.
We evaluate these theoretical claims using newly assembled data on bureaucratic communication and government performance regarding citizen complaints in China. First, we draw upon a rare dataset of internal email communication between city-level officials (patrons) and county-level officials (clients), which allows us to directly observe client–patron communication. Second, we compile several original datasets on how county officials publicly respond to citizen complaints. Our empirical strategy compares county officials who have patronage ties to city leaders—identified through shared birthplace or workplace histories—with otherwise similar officials who lack such ties.
In internal emails, we find evidence of clients engaging in information campaigns. First, client counties communicate more with higher-level officials compared to non-client counties. Second, they are more likely to signal diligence in their resolution of citizen complaints. Finally, they are also more likely to defend themselves in response to citizen complaints of wrongdoing. However, when examining data on actual public responses to citizen complaints, we find that client and non-client counties expend similar levels of effort in responding to citizen concerns. Altogether, clients are more likely than non-clients to talk up their effort and capabilities to their patrons without actually providing more support to the public.
Beyond the scholarship on patron–client linkages, these findings have implications for the literature on authoritarian responsiveness (Malesky and Schuler, Reference Malesky and Schuler2012; Distelhorst and Hou, Reference Distelhorst and Hou2014, Reference Distelhorst and Hou2017; Su and Meng, Reference Zheng and Meng2016; Kornreich, Reference Kornreich2019). First, our results depart from prior research, which emphasized the role of formal institutions (such as party institutions) in shaping nonelectoral responsiveness. While formal institutions are generally found to improve authoritarian responsiveness (Chen, Pan and Xu, Reference Chen, Pan and Yiqing2016), competition for the spoils of patronage degrades government responsiveness to the public. Second, these results provide an additional rationale for how responsiveness can be useful in nonelectoral settings or by nonelected officials. Scholars have argued that responding to the public is valuable in nondemocratic settings because it can assist in state surveillance and information gathering about public opinion and preferences (Meng et al., Reference Meng, Pan and Yang2014; Dimitrov, Reference Dimitrov2014c; Manion, Reference Manion2016) or as a way to alleviate public pressure when actual systematic changes are difficult (Ding, Reference Ding2020). Our results show that responsiveness can also bolster informal ties and patronage relations between bureaucrats.
More broadly, this paper contributes to a growing body of work on political communication between officeholders and the public in nondemocratic settings (Pan and Chen, Reference Pan and Chen2018; Pan, Reference Pan2019; Lu and Pan, Reference Lu and Pan2021; Chang, Reference Chang2021, Reference Chang2024; Chapman, Reference Chapman2021; Lueders, Reference Lueders2022). Our results imply that even though there are functional similarities between government responsiveness in electoral versus nonelectoral systems, these similarities result from very different political motivations. Officials in democracies may be eager to respond to public complaints and demands to improve their chances of reelection. Officials in nondemocracies may be similarly eager to respond to public complaints, but because they can use these activities to curry favor with patrons. Recognizing these differences in political motivations helps us better understand the nature and effects of political communication in nondemocratic settings.
This paper proceeds in four sections. Section 2 lays out our theoretical framework for why clients engage in information campaigns. Section 3 uses the dataset of internal email communications to delve deeply into patron–client relations in one Chinese prefecture, demonstrating how client counties communicate differently with prefecture leaders compared to non-client counties though they expend similar levels of effort in responding to grievances publicly. Section 4 extends our argument to thousands of counties across China using several additional datasets. Section 5 explores the political and social consequences of information campaigns. Section 6 concludes.
2. Information campaigns in patron–client relations
How do social relationships between government officials impact information flows between higher- and lower-level government bureaucracies? A large body of work in political science and economics contends that lower-level agents can have objectives that differ from their principals. The presence of information asymmetries—that is, a lack of communication between principals and agents—allows the agents to pursue their own interests at the expense of the organization as a whole. Subsequently, the literature has been concerned with describing the adverse effects of information asymmetries and offering potential solutions such as increased monitoring of agents. However, we know less about the nature of these information asymmetries and political communication between principals and agents more broadly. What information is shared with principals and what is not? How do lower-level agents communicate with higher-level principals, and what determines the nature of their communication? And what are the consequences of internal communication to governance outcomes?
We attempt to address some of these questions. The central claim of this article is that social relations—in particular, patron–client relationships—lead to increased informal communication between different levels of government hierarchy. Lower-level clients engage in information campaigns to present themselves as diligent, competent government officials. We argue that not all information asymmetries are created equal: when patron–client relations overlap with formal superior-subordinate relations, these social relationships can increase communication between different levels of government as clients engage in information campaigns toward their patrons.
2.1. Patron–client relations
Patron–client relations are voluntary, reciprocal exchanges between individuals of unequal status, whereby the individual with higher status (the patron) provides benefits such as resources or protection to the individual with lower status (the client) in exchange for some service, skills, support, or general assistance (Scott Reference Scott1972). Patron–client relations often also encompass a social dimension: patrons and clients tend to be socially embedded through kinship networks, ethnicity, regionalism, or religion. Because patron–client relationships are inherently social, the nature of these relations is fluid and subject to change. Thus, a key aspect of patron–client relations is maintaining this social relationship.
From the perspective of the client, their priority is to ensure this mutually beneficiary relationship with the patron is sustained. Importantly, the patron–client relationship is not enforced by another third-party but through the mutual interests of maintaining a valuable long-term relationship. In a bureaucracy, the most important good to be distributed is the promise of future promotion. However, there is some uncertainty in whether a client could ultimately receive that good. If the good that is being distributed is limited in nature, the presence of other clients heightens this uncertainty and creates competition among clients. Bureaucratic promotion is one example of a limited good (Hillman, Reference Hillman2010). Not all clients, despite being in the same patronage network, can attain the most desired political benefits, resulting in competition among clients for these high-valued spoils. This context of competition among clients and limited political goods means it is crucial clients work to maintain their relationship with patrons.
So how can clients maintain their relationships with patrons? Clients may work to demonstrate their worthiness, competence, and loyalty through various means. First, clients could help patrons achieve their policy goals. In government bureaucracies, upper-level officials must delegate to subordinates—for example, a top city official cannot attract investment, provide public goods, or maintain social order alone; he needs subordinates to complete these tasks. In the process of implementing these tasks, wrongdoing or neglect of duties by lower-level officials can pose especially serious problems for the political careers of their patrons. For example, wrongdoing may generate public dissatisfaction, contention, or unrest that may impede patrons’ own upward career trajectories or even lead to demotion (Nathan, Reference Nathan2003; Dimitrov, Reference Dimitrov2014a; Qin et al., Reference Qin, Strömberg and Yanhui2015). Simply put, when patron–client relations overlap with formal superordinate–subordinate roles in a government hierarchy, the patron would want the client to fulfill their governance tasks faithfully and competently. To then please their patron, the client could invest effort into achieving the patrons’ policy priorities. In fact, existing research suggests that this may be the case (Jia et al., Reference Jia, Kudamatsu and Seim2015; Jiang, Reference Jiang2018; Toral, Reference Toral2024). Clients could also demonstrate relative competence by undermining other clients (Chen and Hong, Reference Chen and Hong2021).Footnote 1
Building on this literature, we argue that another way clients can maintain relationships with patrons is to engage in information campaigns. We define information campaigns as the communication strategies local officials use to project an image of competence and loyalty to upper-level leaders. This strategy includes (but is not limited to) propagating their governance achievements in local newspapers and other communication channels (e.g., meetings, emails, websites) directed toward their upper-level bosses. Broadly, clients may believe that increased communication with the patron is one way to maintain this vital relationship and affords them opportunities to continually signal competence and loyalty. Clients want to invest in the relationship with their patron and a steady flow of communication between patron and client can ensure that this relationship is sustained.
While existing works on information campaigns have focused on how politicians build a positive image among citizens to influence public opinion (Schoenbach et al., Reference Schoenbach, De Ridder and Lauf2001; Guerrero-Solé, Reference Guerrero-Solé2018; Woolley and Howard, Reference Woolley and Howard2018; Ding, Reference Ding2020; Erlich, Reference Erlich2020), our paper aligns with Cai (Reference Cai2004); Shih (Reference Shih2008b), and Pan (Reference Pan2019) in highlighting a type of information campaign directed toward upper-level leaders within a bureaucracy. This type of information campaign is especially important in understanding the dynamics of bureaucratic patron–client relationships in authoritarian regimes.
Information campaigns are more likely to be present in policy arenas where subordinates can engage in a coordinated response, and there is relatively weak top-down monitoring. This is because coordination ability and weak monitoring offset the two main costs of information campaigns. The first cost of implementing an information campaign is a coordination cost. To execute an information campaign, a subordinate must coordinate with different local departments, agencies, and bureaus, and then present a unified front to their bosses through various communication channels. If coordination costs are very large due to the complexity of a task, information campaigns become less likely. The second cost is the risk of being “caught” by bosses when they discover a discrepancy between what subordinates report and the facts on the ground. When top-down monitoring is weak, the probability of incurring this cost is low. This is especially true if subordinates are not shirking their responsibility completely but devoting some resources to demonstrating actual competence in addition to conducting an information campaign.
2.2. Using citizen complaints in information campaigns
Government responsiveness is an important arena of governance, even in a nondemocratic setting, and superiors care about how their subordinated agencies respond to public complaints. Gathering and responding to public complaints help leaders learn about public opinions and preferences (Meng, Pan and Yang, Reference Meng, Pan and Yang2014; Dimitrov, Reference Dimitrov2014a, Reference Dimitrov2014b, Reference Dimitrov2014c; Manion, Reference Manion2016). Public grievances also allow leaders to detect signs of disruption and dissidence (Pan, Reference Pan2020). In order to encourage the public to submit complaints and allow the regime to gather critical information about public opinion, government agencies must also demonstrate certain levels of responsiveness (Distelhorst, Reference Distelhorst2017). Additionally, failing to respond to public complaints can lead to public unrest and protests, which is detrimental to both clients’ and patrons’ careers in an authoritarian bureaucracy (Edin, Reference Edin2003; Whiting, Reference Whiting, Yang and Naughton2004).
Accordingly, authoritarian governments recognize that one key strategy in maintaining stability is actively responding to public complaints and co-opting dissidents (Dimitrov, Reference Dimitrov2014b, Reference Dimitrov2014c). For example, Chinese leaders have stressed the importance of addressing public complaints, and public responsiveness has become a critical criterion for the performance evaluation of lower-level officials.Footnote 2
While complaints can be made publicly (e.g., posted to the Internet), information asymmetries between superiors and subordinates persist. This is because the higher-level task of responding to grievances is one that consists of many small, but similarly structured tasks such that superiors use general heuristics (e.g., does the subordinate bureau operate a functioning digital platform to collect grievances?) and spot checks (e.g., investigate the resolution of a few complaints) to assess lower-level performance. Bosses cannot investigate subordinates’ handling of every complaint, which means that they do not have perfect information or perfect monitoring ability. There is empirical evidence of this in China, where despite high state capacity, upper-level leaders do not have enough time or labor to check the work of lower-level leaders, and lower-level leaders routinely engage in information manipulation, including manipulation of data related to public grievances (Tsai, Reference Tsai2008; Wallace, Reference Wallace2016; Pan and Chen, Reference Pan and Chen2018; Pan, Reference Pan2019).
Officials can address complaints by investing resources and time to investigate grievances, adjudicate disputes, resolve underlying issues, and communicate outcomes to complainants. Doing so can help avoid major scandals or public unrest that would catch the attention of superiors (Ding, Reference Ding2020). On the other hand, officials can also choose to shirk their duties of responding to complaints. For clients, public grievances is one arena where there are incentives for subordinates to engage in information campaigns due to existing information asymmetries and superiors’ difficulties in fully assessing the performance of subordinates.
2.3. Observable implications
The observable implication of our theoretical framework is that clients will engage in information campaigns to project an image of competence in resolving public grievances. We look for evidence of this in China, where patron–client relationships are a persistent feature of the Chinese bureaucracy that shapes political and economic outcomes. Patron–client relationships play an important role in career advancement (Hillman, Reference Hillman2010; Shih et al., Reference Shih, Adolph and Liu2012; Li and Gore, Reference Hui and Gore2018), local political and economic relationships (Oi, Reference Jean1985), economic outcomes (Jia et al., Reference Jia, Kudamatsu and Seim2015; Jiang, Reference Jiang2018), and information manipulation (Pan and Chen, Reference Pan and Chen2018). Importantly, patron–client relationships in China are often characterized by asymmetry, where patrons control resources that clients desire and cannot obtain without the acquiescence of the patron. Not every client can enjoy the most sought-after spoils: promotions within the bureaucracy. Indeed, an assumption in many works on patron–client relationships within the Chinese bureaucracy is that there is competition for political jobs (Shih et al., Reference Shih, Adolph and Liu2012; Jia et al., Reference Jia, Kudamatsu and Seim2015; Jiang, Reference Jiang2018; Chen and Hong, Reference Chen and Hong2021). Potential clients who wish to curry favor with powerful patrons may use a variety of ways to obtain promotions, including signaling loyalty through mass media (Shih, Reference Shih2008a) or increasing GDP (Jiang, Reference Jiang2018). Lower-level officials engage in these actions because there is a limited number of high-value political positions and not all potential clients can receive a promotion from his or her political patron.
3. Displaying diligence and refuting criticisms
Information campaigns can occur through multiple channels, such as internal email communications or public-facing government websites or social media accounts. Although internal communication and public-facing communication are distinct from one another, both types of communication are crucial for local governments to signal competence and diligence to their superiors.
3.1. Email communication data
Very rarely do we observe routine interactions between patrons and clients. Anthropologists and others who conduct in-depth fieldwork can observe these interactions but usually only for a limited number of patron–client dyads. We leverage a unique dataset that contains email communications between patrons and clients in a Chinese prefecture, which gives us insight into routine and repeated patron–client interactions.
This dataset is a leaked archive of emails sent to and from the Propaganda Department (宣传部) of J. Prefecture from 2012 to 2014 (Henochowicz, Reference Henochowicz2014; Sonnad, Reference Sonnad2014). This archive of emails remains publicly available,Footnote 3 and previous scholars were able to systematize this rich and complex archive (King et al., Reference King, Pan and Roberts2017; Pan and Chen, Reference Pan and Chen2018). Because the main task of local propaganda departments is to monitor and “manage” online public sentiment, almost all of the emails pertain to public complaints and grievances that have received online attention. This includes emails from J. Prefecture’s 13 subordinate counties that describe how county-level governments addressed public complaints and grievances.
We identified 2,768 emails sent to and from the J. Prefecture Propaganda Department between May 14, 2012 and December 6, 2014. Among them, 1,239 describe how government agencies handled public grievances. Because some emails describe more than one grievance, we identified a total of 1,333 complaints and government responses described within this email archive. More than 70% (968) of emails were sent from subordinate counties to the J. Prefecture Propaganda Department.Footnote 4 Many of the email senders were county propaganda department staff who were tasked with reporting the county government’s responsiveness to J. Prefecture Propaganda Department.Footnote 5 J. Propaganda Department then informed J. Prefecture’s mayor and Party standing committee members about how their subordinate counties addressed online grievances (Pan and Chen, Reference Pan and Chen2018).
In these emails, some county officials only copy and paste the text of how they responded to public complaints and grievances (response), while others also provide prefecture-level officials with additional information on how they handled the grievance (commentary). For example, one email from YX county to J. Prefecture is titled “Regarding a netizen raising the issue of no street lights in TZ Garden.” This email only contains a response, which is simply the answer this county bureau gave to the public grievance, written below.
Dear Netizen: Hello! Regarding the problem you raised of TZ Garden having no street lights, the County Street Light Management Office sent people to that place to investigate. The outer road near the entrance/exit of TZ Garden is about 100 meters long and 6 meters wide. This road connects GH Road and KF road. For the convenience of the public’s evening travel, our bureau will install 3 street lights on this section of the road in the near future.
However, this email does not contain any additional commentary directed toward upper-level officials.
Other emails contain additional information sent to prefecture-level officials about how the county government handled the public grievance. For example, one email from LS county to J. Prefecture is titled “Explanation of the criticism ‘This building can be called J. Prefecture’s Dirtiest Building by LS District.’” The email begins by quoting the original complaint:
This building can be called “J. Prefecture’s Dirtiest Building” and seriously affects the image of the city. I recommend the facade be cleaned, and preferably be built as a full glass wall so it can be linked in the future with the Jiulong Plaza that J. Prefecture wants to develop. [The government] must pay attention to its image.
The commentary then goes on to explain what LS county has done to address the complaint:
After we received this complaint, we immediately reported it to the county leaders, and at the same time, contacted relevant departments to better understand the matter. Because the building belongs to the prefecture Bureau of Commerce, there is currently no department in LS County that can take charge of the issue. Therefore, we want to explain to you, the prefecture leaders, that we are unable to resolve this issue. Thank you!
This email from a county to the prefecture essentially communicates: we know we received this very nasty complaint online. We took it seriously and expended effort to look into it, but we do not have the power to resolve this issue because the building is controlled by the prefecture (you). We want to let you know there is nothing we can do.
This rare look at internal, informal communications between subordinate county governments and their upper-level prefecture government allows us to explore relationships among government actors in the context of public responsiveness and to compare whether patterns of upward communication systematically differ between clients and non-clients. Beyond the main analyses, we also use communications from public-facing websites. Since our unit of observation is the county, in the subsequent section, we refer to clients as client counties and non-clients as non-client counties.
3.2. Measuring patron–client ties
We draw from existing research on patronage in China that use birthplace and workplace affinities to measure patron–client relationships (Burns, Reference Burns1994; Cheng, Reference Cheng2000; Shih et al., Reference Shih, Shan and Liu2010; Choi, Reference Choi2012; Shih et al., Reference Shih, Adolph and Liu2012; Hillman, Reference Hillman2014). We define a county as a client county if two or more prefecture party standing committee members were either born or worked in that county. In this sense, we depart from Shih et al. (Reference Shih, Adolph and Liu2012), which coded patron–client relationships as ties between individuals (i.e., relationships between individual elite Chinese Communist Party members).
The logic behind this coding is that if a prefecture party standing committee member previously worked in a particular county, he or she would have many contacts in the county government who are also connected to the county mayor, party secretary, and the leader of the propaganda bureau. Though we may not be able to directly measure patron–client ties, we assume that these counties will be socially closer to the prefecture party standing committee member compared to other counties. This patron–client coding rule was applied to all the datasets used in this paper for consistency.Footnote 6
We collected the biographical information of prefecture party standing committee members who were in office during the period of our email communication data.Footnote 7 Identifying prefecture party standing committee members was a significant challenge. We used data from prefecture yearbooks and the Internet Archive, which provides access to historical snapshots of Chinese government websites, to identify these individuals. For some leadership positions, such as military positions, biographical information is particularly difficult to find. In addition to prefecture yearbooks and websites, we also used guqiaowang, a commercial public forum of leader biographies.Footnote 8
Using our definition of patron–client relations, we identified 6 (out of 13) counties as client counties of J. Prefecture.
3.3. No differences in effort between clients and non-clients
We first examine the level of effort expended by county governments to address public grievances. We find two types of grievances in the emails: complaints that can be solved by the county simply by providing information, and those that can only be solved if the county devotes resources beyond a written response. The level of effort required to resolve the first type of grievance is always lower, but within each type of complaint, the level of effort county governments expend can vary. We create three variables to capture the variation in effort required to address these two types of complaints.
The first variable describes whether or not (1 or 0) the county provides a helpful response (Helpful Response) in answering a complaint that can be solved by providing information. We measure Helpful Response by having a team of trained research assistants code the response contained in the email (not the commentary). This response to a complaint about how costly it is to get a marriage certification is an example of a helpful (high effort) response.
“Is it reasonable that the DC county Civil Affairs Bureau charges 20RMB to certify that one is single [not married]?” Thanks for the netizen’s concern. According to the netizen’s post, the Civil Affairs Bureau charges 20RMB to certify that a person is single. To issue this certificate, we must investigate an individual’s dossier, and only after we verify that the person is not married can we issue the certificate. The 20RMB is the verification fee, and the details of this fee can be found in the document issued by the County Price Bureau “Dujia Fuzi (2012) No.11” as the fifth item in the document.
A response that is not helpful (low effort) would be one that told the complainant to contact someone else, or simply stated 20 RMB is reasonable without providing any additional explanation.
The second variable captures the length of the government response. As Blumenstock (Reference Blumenstock2008) suggests, the length of content can be an effective indication of its quality. Scholars have also used length as an indicator for quality when measuring government documents (Gratton et al., Reference Gratton, Guiso, Michelacci and Morelli2019). In our context, longer responses are more likely to contain useful, pertinent, and detailed information, and more likely to describe on-the-ground complexities. We calculate the length of government responses in three different ways. First, we calculate the number of Chinese characters in a government response. Second, we count the number of Chinese characters, excluding punctuation. Third, we calculate the number of words (which could be made up of one or more characters) in a government response.
The third variable describes whether or not the county has taken action (Take Action)—expended human or other resources—to investigate or resolve the complaint. For this variable, research assistants again code the response contained in the email (not the commentary). The following response is an example of the government taking action (high effort):
Response to “Some people ‘diverted the river to create a pond’ causing hidden dangers on the south side of the Weihe bridge in YX County”: Hello! Now we reply to your post on the forum of J Province: in September 2014, the water administrative department of our bureau issued a notice of suspension in accordance with relevant laws and regulations to the relevant person LZS, and requested that he restore the pond to its original condition within a time limit. Our bureau investigates water violations every time we discover them and no violation will be tolerated. Thank you again for your supervision and support of our work! YX County Water Affairs Bureau December 5, 2014.
An example of a response where the government does not take action (low effort) is:
At present, there are two medical emergency telephones in RC, County 120 and 920. The system is not confusing. According to the netizen’s suggestion, the RC County Health Management Department will improve the situation if the actual situation requires it. Thanks to the netizen for your concern for RC Health.
In coding government actions, we are taking the county governments at their word. We assume that none of the public officials are lying about what actions they have taken. We believe this is reasonable because counties can increase the perception of effort without taking the risk of lying outright.
We find no large or statistically significant differences between client and non-client counties in the helpfulness of their responses to complaints, in the length of their responses to complaints, and in the frequency with which they take action (expend resources) to address grievances. The three rows in Table 1 show the results of a Welch two-sample t-test for each of the three variables.
Comparison of effort between client and non-client counties in J. Prefecture
Table 1 Long description
Measures compare client versus non-client counties on helpful response rate, average response length in words, and take-action rate, each with a reported p-value. Helpful responses are 0.95 for client counties and 0.94 for non-client counties, with p-value 0.74. Average response length is 211.93 words for client counties and 206.30 words for non-client counties, with p-value 0.68. Take-action rates are 0.77 for client counties and 0.79 for non-client counties, with p-value 0.36. Across all three measures, the values are very close between groups and the p-values suggest no statistically meaningful differences. Results summarize group averages and do not indicate sample size or variability, so small differences should be interpreted cautiously.
We find that 95% of the responses in both client and non-client counties are Helpful Responses. The length of responses in client counties is slightly longer (212 words on average) than in non-client counties (206 words), but this difference is not statistically significant. We find that client counties’ responses show the government taking action (expending resources in either investigating or resolving the complaint) 77% of the time. While this proportion is slightly higher in non-client counties (79%), the difference is not statistically significant. These results are robust when we control for characteristics of the grievances (see Supplementary Appendix 2).
3.4. Engaging in information campaigns
Next, we examine whether or not client counties engage in information campaigns—attempting to increase the perception of their effort and capabilities in their emails to the prefecture (patron). We create three variables to test this claim. First, we examine the volume of emails between counties and the prefecture. We would expect client counties to email the prefecture more frequently to demonstrate constant and high levels of effort. We find that this is the case. The six client counties sent a total of 641 emails while the seven non-client counties sent a total of 326 emails to the prefecture in the same time period.
Second, we look at whether there is additional commentary in the email beyond the response that was posted publicly, and whether this commentary highlights how diligent (Displaying Diligence) the county was in resolving the complaint. For example, in an email from one county government to the prefecture about a fire at a steel factory, the county not only attached its response to the complainant, it also explained to prefecture leaders that the fire had not generated negative effects on public welfare or public opinion.
Focusing on the commentary helps us capture manipulation of perceived effort by the county because the commentary is never seen by the public. Presumably, if the information was valuable to the public, the county would have included it in its public response. The recipient of this kind of internal information is clearly the prefecture government, not the greater public.
We find that client counties are more likely to display diligence by including additional commentary to superiors. This result is substantively and statistically significant. Client status is associated with a 2.06 increase in the odds ratio of displaying diligence through additional commentary when communicating to the prefecture about public grievances compared to non-clients. This result holds, even when we control for variables that may provide alternative explanations for the difference we observe. These results are presented in Table 2.
Displaying diligence
Table 2 Long description
Regression results across five model specifications report coefficients with standard errors in parentheses and sample sizes near 770 to 970. Client status is positive in every model and is statistically reliable in all five, ranging from 0.262 in model 4 to 0.721 in model 5. Mayor tenure year is negative and statistically reliable where included, at minus 0.695 in model 2 and minus 1.484 in model 5. In model 5, several agency categories are negative and statistically reliable: public security (minus 1.352), village government (minus 1.269), and other government agency (minus 1.544), while propaganda is negative but not statistically reliable. County population and county GDP per capita are not statistically reliable in model 3 or model 5. In model 4, reliable is negative and statistically reliable (minus 0.475), while individual grievance is positive with weaker statistical support (0.327). Constants vary widely across models, suggesting baseline differences across specifications; results should be interpreted as associations that depend on included controls and changing sample sizes.
$^{***}p$ 
$ \lt $ 0.01; 
$^{**}p$ 
$ \lt $ 0.05; 
$^{*}p$ 
$ \lt $ 0.1.
All models in Table 2 are logistic regressions.Footnote 9 Column (2) includes controls for county leader characteristics, which may also influence how county officials communicate with their superiors. These include the county mayor’s gender (Mayor gender) and his/her year in office (Mayor tenure), which has been shown to influence image building and online communication among county leaders (Pan, Reference Pan2019).Footnote 10 Column (3) includes controls for county-level characteristics that may affect public grievances and by extension responses to them. These include the log population of the county, and the county’s log GDP per capita.Footnote 11 Column (4) includes grievance characteristics that may also influence the language used by county governments in their responses. These include whether a grievance pertains to an individual or household rather than affecting a larger group (Individual grievance); whether a complaint is within the jurisdiction of the county to solve (County jurisdiction); whether the complaint is reliable as determined by whether it is based on a person’s personal experience (Reliable); and whether the grievance requires that the county government expend resources to resolve—such as conducting an investigation, providing additional public goods, changing an existing policy, or mediating disputes (Resource intensive). Column (4) also includes a variable we extract from the text of the email that captures the style of communication between counties and the prefecture because displays of diligence could simply reflect different norms of communication. We identify whether emails use terms of respect for the prefecture and place themselves as subordinate to the prefecture in the email as a proxy of communication norms. Column (5) includes controls about the government bureau or agency that responded to the public grievance.Footnote 12 The difference in communication that we observe may simply reflect differences among various government agencies. Primary response agencies are county-level governments, village-level governments, public security agencies (i.e., police bureaus, police stations, and traffic police), and subordinate propaganda departments. Other responding agencies include education bureaus, schools, hospitals, family planning bureaus, and social welfare bureaus, to name a few. More detailed information about the response agencies is included in Supplementary Appendix 1. Client status remains statistically significant, even when controlling for who wrote the public response. Across all of these specifications, being a client county increases the likelihood that an email to the prefecture displays diligence.
Another strategy counties can take to increase the perception of effort and capability is to refute criticisms by denying the veracity of complaints (Denial), especially those that reflect badly on their performance. For example, with regard to a public complaint that members of the county meteorological department were forced to attend the funeral of the father of the department’s boss, the county tells the prefecture that employees voluntarily took time off, and no public resources were used to take employees to the funeral. In another example, a complaint states that a village school in HK county forced students to eat (and pay for) school meals. The county denies the veracity of these complaints and tells the prefecture that it charged the appropriate meal fees and did not force students to eat. Sometimes these refutations are in the actual response to the public, but many other times, they are found in the commentary.
We find that client counties are more likely to refute criticisms compared to non-client counties. Table 3 provides estimates of logistic regressions and shows that client status positively predicts denying the veracity of public complaints.Footnote 13 This result is substantively and statistically significant. In the full model with all control variables (Model 5), client status is associated with a 1.46 increase in the odds ratio of denying the veracity of public complaints.
Denying veracity of public complaints
Table 3 Long description
Regression results across five model specifications estimate predictors of denying the veracity of public complaints, reporting coefficients with standard errors in parentheses and sample sizes per model. Client status is positive and statistically reliable in every model, ranging from about 0.37 to 0.50, indicating higher likelihood of denial for clients. Adding mayor characteristics in model 2 does not change this pattern; mayor tenure year and mayor gender are not statistically reliable there. County economic controls in model 3 show a negative, statistically reliable association for county gross domestic product per capita (about minus 0.90), while county population is negative but not statistically reliable. Complaint and organizational features in model 4 highlight resource intensive cases (about 1.24) and norm of communication (about 0.72) as positive and statistically reliable; other included factors are not clearly distinguishable from zero. In the fullest model 5, resource intensive remains strongly positive (about 1.28), village government (about 0.91) and other government agency (about 0.58) are also positive and statistically reliable, while public security and propaganda are not. Sample sizes vary by model from 768 to 968, so comparisons across columns should note that different covariate sets use different numbers of observations.
$^{***}p$ 
$ \lt $ 0.01; 
$^{**}p$ 
$ \lt $ 0.05; 
$^{*}p$ 
$ \lt $ 0.1.
For ease of interpretation, in Figure 1 we plot the predicted probability that an email from a non-client and a client county displays diligence or contains a denial of wrongdoing, using the results from Column 1 of Table 2 and Column 1 of Table 3. Client counties are 6 percentage points more likely to display diligence in their communication to upper-level officials and 9 percentage points more likely to deny wrongdoing compared to non-client counties.
Predicted probabilities of diligence and denial.
Overall, these results illustrate that within J. Prefecture, client counties attempt to appear as if they are devoting more effort and are more capable of responding to public grievances. They do this mainly through displaying diligence and denying the veracity of public complaints in information campaigns aimed at their patrons. However, we do not observe actual differences between client and non-client counties in the helpfulness or quality of their responses to the public.
4. Information campaigns across China
Beyond internal communications, public websites are also key venues where local governments can signal competence to their superiors. We used existing and novel data of public communication to uncover two core results that echo our findings from internal communications data. First, client and non-client counties do not differ in the actual level of effort expended in answering public citizen complaints. Secondly, client counties, however, differ in how they indirectly shape their superiors’ perceptions of their diligent attentiveness to public complaints.
4.1. No differences in effort across China
Within J. Prefecture, we evaluate the subordinate government’s effort based on the content of county responses to public complaints. We can do the same for other counties across China. We collect data on government responses to public complaints and code whether the responses are helpful (Helpful Response), measure the length of responses, and code whether the response shows the government taking action (Take Action), consistent with the measures we used to examine responses in the internal emails.
We analyzed two datasets of government responses. The first comes from Chen et al. (Reference Chen, Pan and Yiqing2016)’s field experiment on authoritarian responsiveness. In this experiment, citizen complaints were submitted to the online complaint forums of all counties in China with such online forums. All complaints were requests for assistance in obtaining dibao, a non-conditional cash transfer program. The type of appeal written in the information request was randomized, and one of the appeals included a threat to report the county government to upper-level leaders if the complaint was not resolved. This treatment threatens potential monitoring by upper-level leaders, so we analyze the responses from client and non-client counties in this treatment arm. The advantage of this dataset is that it captures both publicly posted responses by county governments and private responses (e.g., delivery via private email or response posted online but only available when logged in). The disadvantage of this dataset is that these are fictitious requests and are all focused on one issue.
The second is a novel dataset we collected of county government responses to public complaints posted on prefecture government websites. County governments not only write responses to citizen complaints on their own government websites, but they also answer citizen complaints hosted on prefecture websites that are directed toward them. Because prefecture governments host and maintain these websites, county responses to grievances posted on prefecture websites are directly observable by the prefecture government. We collect a new dataset of 19,372 citizen complaints and corresponding government responses from 29 prefecture government websites representing diverse regions of China.Footnote 14 We focus on the content of responses written by county governments.Footnote 15
Table 4 shows the results of a Welch two-sample difference in means t-test for each of the three variables in the two datasets that measure responsiveness.
Comparison of effort between client and non-client counties across China
Table 4 Long description
County-level response effort is compared between client and non-client counties using three measures: helpful response rate, response length in words, and take-action rate. For responses to Chen, Pan, Xu (2016), client counties report helpfulness 0.71 versus 0.64, length 123.60 versus 114.25 words, and take action 0.42 versus 0.49, with p-values 0.32, 0.55, and 0.38. For responses on prefecture web forums, client counties report helpfulness 0.99 versus 0.98, length 146.43 versus 132.18 words, and take action 0.73 versus 0.73, with p-values 0.28, 0.26, and 0.94. Across both sections, differences between client and non-client counties are small and inconsistent in direction. The reported p-values are all well above common thresholds, so the table does not provide evidence of statistically meaningful differences between the two groups on these measures.
The results are consistent across the datasets. Client counties across China have a slightly higher proportion of helpful responses than non-client counties, slightly longer responses to complaints than non-client counties, and a slightly lower proportion of responses where they indicate that actual effort has been expended. However, none of these differences are statistically significant.
4.2. Differences in public communication between clients and non-clients
Though internal email communications reveal directly how client and non-client governments communicate with their superiors, other communication channels are also important venues of information campaigns. Specifically, government websites can also be used to conduct information campaigns and present certain “images” to superiors.
First, client counties might influence superiors’ perceptions by publicly and frequently extolling their attentiveness toward the public. Pan (Reference Pan2019) shows how county government websites are used by county officials to build their image, which may or may not reflect their actual performance. One image that county officials project using local websites is one of a responsive and benevolent leader. There is also evidence that higher-level officials examine local government websites (Pan, Reference Pan2019). Client counties may then engage in more public displays of attentiveness toward the public in order to strengthen superiors’ perceptions of their effort to aid the public. We generate a novel dataset of public attentiveness toward the public by using the Google Custom Search API to obtain the count of terms referring to the public (e.g., the public 大众, citizen 市民) from nearly 2,000 county websites in 2014.
We then normalize these counts by taking into account the size of the county government website. We proxy the size of the website by counting, again through the Google Custom Search API, the number of times the county’s name appears on that county’s website. Our theory predicts that client counties will engage in more public displays of attentiveness toward the public than non-client counties. This mimics the measure of diligence in the internal emails because the presence of these key terms projects an image of the county government as a caring and responsive ruler. It does not measure actual responsiveness toward the public.
Our main dependent variables are the sum of each individual key term, the sum of all key terms, and the average of all key terms, normalized by the number of times the county’s name appears on the website. This normalization is done to ensure that our dependent variables are not simply measuring the length or the depth of the county website. Table 5 displays the results of the OLS models with two dependent variables—the appearance of the words “the public” and the appearance of the word “citizen,” and shows that client counties are more likely to talk about “the public” and “citizen” than non-client counties.
Attentiveness toward the public
Table 5 Long description
Regression results estimate how client status and local characteristics relate to attentiveness, reported across six model specifications split by whether “the public” or “citizen” is the reference term. Client status is positive in every model and statistically reliable in most: about 0.02 to 0.01 in the “public” models and about 0.44 to 0.89 in the “citizen” models, with the largest estimate in the fully controlled “citizen” model. Adding controls reduces the client status estimate in the “public” set and increases it in the “citizen” set. Tenure year is near zero and not reliable in the “public” models, but positive and statistically reliable in the “citizen” models once controls are included. Log GDP, log population, and fiscal revenue show mixed, mostly non-significant associations, except fiscal revenue which is positive and statistically reliable in the most controlled “citizen” model. Prefecture fixed effects are included only in models 3 and 6. Sample size is 1,837 in models 1 and 4 and 1,278 in models 2, 3, 5, and 6; coefficients should be interpreted as associations rather than causal effects.
$^{***}p \lt 0.01;\ ^{**}p \lt 0.05;\ ^{*}p \lt 0.1$.
Columns (1) and (4) regress each dependent variable on client status. In columns (2) and (5), we add county-level control variables, and in columns (3) and (6), we include prefecture-fixed effects. This analysis suggests that client counties are more likely to declare their attentiveness to the public on government websites than non-client counties.
Second, counties can influence superiors’ perceptions of their effort by making their responses (or select responses) to public grievances easy for superiors to find. Since many complaints are made online, county governments can increase the accessibility of county complaint forums. We use data from Chen, Pan and Xu (Reference Chen, Pan and Yiqing2016) to capture the accessibility of county web forums in 2014. Specifically, we examine whether county complaint forums make the titles of public posts publicly viewable (defined as visible without logging in), whether they make the bodies of public posts publicly viewable, and whether they make government responses publicly viewable. More accessible government websites ensure that when upper-level officials (or their representatives) visit county-level websites, they can more easily monitor that counties are being responsive. Based on our theory, we would expect client counties to have more publicly accessible websites.
As Table 6 shows, this is what we observe. While a very large proportion of counties make post titles, post content, and government responses public, we observe that a greater proportion of client counties’ public forums are easily accessible in terms of the titles of posts, the content of posts, and government responses than non-client counties. All of these differences are statistically significant at the 0.05 or 0.1 level for a Welch two-sample t-test.
Proportion of county websites with publicly viewable content
Table 6 Long description
Proportions of county websites with publicly viewable content are compared between client and non-client counties across three items: post title, post content, and government response. Client counties are higher for all three measures: post title is 0.93 versus 0.90, post content is 0.91 versus 0.88, and government response is 0.90 versus 0.87. The differences for post title and post content are statistically significant based on p-values of 0.02 and 0.03. The difference for government response is smaller and not statistically significant, with a p-value of 0.08. Overall, both groups show high public visibility, with clients consistently but modestly higher. P-values indicate evidence of a difference but do not convey the size or practical importance of the gap.
5. The political and social implications of information campaigns
Our analysis so far has shown that client counties are more likely to engage in information campaigns, which the literature suggests function less as policy tools and more as signals aimed at maintaining social ties with higher-level patrons. Yet, whether this type of communication carries tangible consequences remains an open empirical question. If information campaigns primarily serve to reinforce social closeness within patron–client networks, do they nonetheless influence political careers or generate observable social benefits for local residents?
To clarify the broader significance of information campaigns, this section evaluates their downstream effects on two outcomes: the promotion prospects of county mayors and the social welfare delivered to citizens. Assessing these consequences allows us to understand whether information campaigns operate mainly as symbolic signals within political relationships or whether they also shape substantive governance outcomes.
5.1. Information campaigns and promotions
First, we examined whether information campaigns—such as displaying diligence or refuting public criticism—were associated with county mayor promotions within J. Prefecture. As shown in Table 7, these activities do not have a statistically significant relationship with the likelihood of county mayors getting promoted.
Information campaign and promotion of county mayor in J. Prefecture
Table 7 Long description
Regression results for promotion are reported in two model specifications with 559 observations each. Denial appears only in model 1 and is near zero (about 0.004) with a small standard error, suggesting little association. Diligence appears only in model 2 and is also near zero (about minus 0.019) with a small standard error. Across both models, client status is strongly negative (about minus 1.02), and mayor age is negative (about minus 0.13), while mayor gender is strongly positive (about 2.23). GDP per capita is strongly positive (about 2.21), fiscal revenue is strongly negative (about minus 0.57), and public expenditure is positive (about 0.28). The constant is large and negative (about minus 16.7). Coefficients and standard errors are very similar across the two models, indicating the main relationships are robust to swapping denial for diligence; statistical significance is indicated by stars, but the table does not establish causality.
$^{***}p$ 
$ \lt $ 0.01; 
$^{**}p$ 
$ \lt $ 0.05; 
$^{*}p$ 
$ \lt $ 0.1.
We then examined whether client counties are more likely to see their mayors promoted than non-client counties using ordinary least squares regressions. Across all counties in China (Table 8), we found that mayors from client counties are 4.5 percentage points more likely to be promoted, even when controlling for leader and county-level variables.Footnote 16 This result indicates that information campaigns by themselves do not directly translate into career advancement. Rather, political ties are more important in determining promotions. Seen in this context, information campaigns appear to function less as an independent path toward promotion and more as a mechanism for maintaining favorable relationships with patrons. Maintaining patron–client relations, however, does have a positive relationship with promotion outcomes.
Client status and promotion of county mayors
Table 8 Long description
Regression results relate promotion to client status and mayor and county characteristics across three model specifications. Client status is positive and statistically meaningful in every model, with coefficients about 0.038, 0.037, and 0.056 as controls expand. Adding mayor traits shows age and ethnicity are positively associated with promotion, while gender is negative and becomes statistically meaningful in the fullest model. In the full model, larger population and higher GDP per capita are both strongly positive predictors of promotion. Fiscal revenue, public expenditure, and GDP growth are not statistically distinguishable from zero in the full model. Sample size declines across models from 2,672 to 2,296 to 1,433, so comparisons should note the changing set of observations.
$^{***}p$ 
$ \lt $ 0.01; 
$^{**}p$ 
$ \lt $ 0.05; 
$^{*}p$ 
$ \lt $ 0.1.
5.2. Information campaign and social welfare
Recent works have demonstrated that patron–client relationships can lead to positive outcomes for citizens (Jiang, Reference Jiang2018). To evaluate whether citizens in client counties in our data benefit from these ties, we conducted regressions using two variables from our county-level dataset: GDP per capita and county government spending on social welfare. Across all counties in China (Table 9), we find that client counties do not differ significantly from non-client counties in overall economic well-being, as measured by GDP per capita (
$p$ = 0.521). Nor do client counties exhibit higher levels of social welfare spending (
$p$ = 0.296).
Client status and social outcomes
Table 9 Long description
Regression results relate client status, fiscal revenue, and logged population to two outcomes: logged GDP per capita and a social welfare measure. Fiscal revenue has a large positive association with logged GDP per capita (about 1,557) and a negative association with social welfare (about minus 0.210), both statistically strong. Client status has an imprecise positive estimate for logged GDP per capita (about 207) and a negative estimate for social welfare (about minus 0.660), with large standard errors, so the direction is uncertain. Logged population is slightly negative for logged GDP per capita (about minus 0.017) and slightly positive for social welfare (about 0.00003), both statistically strong despite small magnitudes. Intercepts are positive in both models (about 21,193 for logged GDP per capita and 2.005 for social welfare). Sample sizes are 1,794 for the GDP model and 1,758 for the social welfare model. Coefficients describe associations and should not be read as causal effects without additional design details.
$^{***}p$ 
$ \lt $ 0.01; 
$^{**}p$ 
$ \lt $ 0.05; 
$^{*}p$ 
$ \lt $ 0.1.
For counties in J. Prefecture specifically (Table 10), we find that refuting criticism is related to a decrease in GDP per capita and there is no relationship between displaying diligence and social welfare spending or GDP per capita. In short, neither clientelist ties nor information campaigns appear to produce measurable social welfare benefits for citizens. Instead, information campaigns are better understood as political signals aimed at maintaining favorable relationships with patrons. This is in contrast to recent literature (Jiang, Reference Jiang2018; Toral, Reference Toral2024) that points out the positive effects patronage can sometimes have on GDP growth and bureaucratic efficiency.
Information campaign and social implication
Table 10 Long description
Regression results report coefficients and standard errors for two outcomes, social welfare and logged GDP per capita, estimated in two model variants. For social welfare, client status is strongly positive at about 2.35 to 2.39, and population (log) is strongly positive at about 16.46 in both variants. Fiscal revenue is also strongly positive for social welfare at about 1.46 to 1.49. For logged GDP per capita, client status is strongly positive at about 0.19, while population (log) is strongly negative at about minus 1.20 in both variants. Denial is negative for both outcomes and reaches statistical significance only for logged GDP per capita (about minus 0.03), while diligence is negative but not statistically significant for either outcome. Sample size is 771 in all models, and constants differ by outcome, so comparisons should focus on coefficients rather than intercepts.
$^{***}p$ 
$ \lt $ 0.01; 
$^{**}p$ 
$ \lt $ 0.05; 
$^{*}p$ 
$ \lt $ 0.1.
6. Conclusion
The literature on patronage and clientelism has mainly concerned itself with the ability of patrons to monitor clients in contexts of electoral clientelism, spurred by the notion that patrons and clients exist in principal–agent relationships with significant information asymmetries.Footnote 17 Our paper asks about the nature of information flow between patrons and clients, and specifically calls attention to the efforts of clients in maintaining this relationship through information flows. We provide a theoretical framework to explain how clients maintain relationships with their patrons through information campaigns. We use a variety of novel datasets, including leaked internal communications between bureaucrats, to test the observable implications of this framework. Our results show that client counties across China are more likely than non-client counties to engage in information campaigns directed at their superiors to shape perceptions about their responsiveness to the public. These information strategies in internal email communications provide additional context to clients’ actions, emphasizing their diligence in responding to citizen grievances and supposed innocence in the face of citizen accusations. On public-facing forums, client counties make their responses to citizen complaints more accessible to upper-level leaders and refer more frequently to the “people” on their public websites. However, client counties do not differ from non-client counties in their actual responses to public complaints. More specifically, they do not differ in the actual helpfulness of their responses, the quality of their responses, or the amount of resources they claim to have expended in addressing the public grievance.
We see this research as the first step in improving our understanding of the dynamics and consequences of how patrons and clients communicate. Many works on patronage focus on resolving the principal–agent problem between patrons and clients through mechanisms designed to monitor and sanction clients (Stokes, Reference Stokes2005; Baldwin, Reference Baldwin2013; Golden and Min, Reference Golden and Min2013). Other works have asserted that patron–client relationships can align the goals of bureaucrats, increase communication between bureaucratic hierarchies, and even lead to improved governance outcomes (Jiang, Reference Jiang2018; Toral, Reference Toral2024). Our work investigates these claims by examining the nature of communication between patrons and clients in a hierarchical bureaucracy. We demonstrate how clients proactively send signals to maintain their relationship with the patron, bringing additional nuance to our understanding of patron–client relations. Our paper joins a small but growing body of work that observes patron–client relationships from the perspective of clients to investigate the communication, political, and economic strategies clients use to maintain their relationship with patrons (Nichter and Peress, Reference Nichter and Peress2017; Jiang, Reference Jiang2018; Pan, Reference Pan2018; Chen and Hong, Reference Chen and Hong2021).
The findings of this paper also provide an alternative explanation for government responsiveness in authoritarian contexts. Consistent with a growing literature in China and elsewhere, we show that responsiveness to the public may be driven by political incentives for career advancement created by formal institutions (Malesky and Schuler, Reference Malesky and Schuler2012; Distelhorst and Hou, Reference Distelhorst and Hou2014; Manion, Reference Manion2014; Meng et al., Reference Meng, Pan and Yang2014; Distelhorst, Reference Distelhorst2017). We depart from this literature, however, by highlighting the role that government responsiveness plays in informal, patron–client relationships that are prevalent in the Chinese bureaucracy. Responsiveness to the public is used as an avenue to signal positive attributes of client counties to patrons. While client counties portray themselves as more responsive to their superiors, they do not perform better than non-client counties in actual metrics of responsiveness to citizens.
Finally, our findings shed light on information manipulation in bureaucracies. Researchers often examine information manipulation as occurring between government and society and through avenues such as censorship (King, Pan and Roberts, Reference King, Pan and Roberts2013), falsification (Wallace, Reference Wallace2016), fabrication (Roberts, Reference Roberts2018), or concealment (Pan and Chen, Reference Pan and Chen2018). Here, information manipulation occurs within the bureaucracy through the manipulation of perception. Although this form of manipulation is subtle and does not directly involve the public, it undermines patrons’ knowledge about their clients’ performance, which exacerbates the dictator’s information dilemma concerning the actions of subordinates and the distribution of political spoils.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2026.10112. To obtain replication material for this article, https://doi.org/10.7910/DVN/2DWBAK.
Acknowledgements
We thank Jennifer Gandhi, Noah Nathan, Jean Oi, Scott Rozelle, and Andrew Walder for their extremely helpful comments and suggestions. We are especially grateful to Jennifer Pan for her generosity in sharing data and for her insightful comments, particularly at the beginning of this project, which were instrumental in bringing this project to completion. Zhiheng Xu, Xueying Yan, and Shuxin Meng provided excellent research assistance. Finally, we want to thank the editors and anonymous reviewers who greatly improved every aspect of this paper.
Funding statement
This work was supported by the National Science Foundation (grant 1831481) and the Stanford China Fund.
Competing interests
The authors report that there are no competing interests to disclose.
Open access
