Journal of Development Economics 160 (2023) 102985

Contents lists available at ScienceDirect

Journal of Development Economics

journal homepage: www.elsevier.com/locate/devec

Regular article

The political impacts of land expropriation in China✩

Wenbiao Sha
Lingnan College, Sun Yat–sen University, Guangzhou, China

ARTICLE INFO ABSTRACT

Keywords: I study the political consequences of state expropriation of agricultural land in rural China by using national
Land expropriation individual-level panel data for the period 2010–2018. Comparing outcomes before and after expropriation
Political consequences with changes among individuals not experiencing expropriation, I find that having one’s land expropriated
Governance
decreases individuals’ trust towards local government officials, and increases the incidence of having conflicts
Development benefits
with local government officials. I also provide evidence that the adverse political impacts can be mitigated by
better local governance, undertaking projects with public benefits, and ex-ante non-agricultural employment.

1. Introduction public goods and to develop the economy more generally.3 Since these
goals deliver benefits to the public, whether expropriation has political
State expropriation (also known as ‘‘eminent domain’’) is both costs, and if so, how detrimental the costs are, could be context-
common in much of human history and prevalent in today’s developing dependent. Public perceptions of the motivations for expropriation
countries.1 Economists have long been interested in the consequences could influence the extent to which expropriation leads to adverse
of state expropriation. Unconstrained executive power may lead the political impacts. To explore this relationship, this study examines two
government to extract rents from citizens and to deprive households circumstances that may enable the state to implement expropriation
and firms of secure property rights, especially in the developing world projects without losing citizens’ political support. First, if local officials
where oftentimes accountability is limited and corruption is pervasive. have a reputation for good governance, the expropriation projects are
In such contexts, state expropriation often has unfavorable outcomes. likely to be well managed and justified and the process of expropriation
Scholars have provided rich empirical evidence on its economic costs,2 may be implemented in a manner which is more transparent. In this
but little evidence is available on its political consequences. In this case, state expropriation may not necessarily have high political costs.
paper, I study to what extent state expropriation may have political
Second, public use of the expropriated resources may be important.
costs, specifically whether citizens subject to expropriation lose trust
When the projects generate public benefits, the interests of both the
in, or have more conflicts with, public officials.
government and the citizens are aligned and adverse political impacts
State expropriation is generally justified by the public interest. The
may not emerge.
state is expected to employ expropriation as a policy tool to provide

✩ I am deeply indebted to my supervisor Albert Park for his advice and support throughout this project. I am grateful to the editor Gerard Padró-i-Miquel,
two anonymous referees, Maitreesh Ghatak, Meixin Guo, Li Han, Guojun He, Ning He, Jean Hong, Ruixue Jia, James Kung, Edward Miguel, Wenwei Peng,
Sujata Visaria, Jin Wang, Shaoda Wang, Noam Yuchtman, Y. Jane Zhang, and seminar and conference participants at Berkeley Development Lunch, Berkeley
HSSA Symposium, HKUST Empirical Development Seminar, HKUST Applied Micro Seminar, the 12th World Congress of Econometric Society, the 2021 China
Meeting of the Econometric Society, the 4th Annual Conference of Government and Economics at Tsinghua ACCEPT, and the 5th International Conference of
China Development Studies at SJTU for helpful comments and suggestions. All errors are my own.
E-mail address: [email protected].
1
Throughout this paper, I use state expropriation and government expropriation interchangeably, referring mainly to the power of the state (government at
various levels) to take land property from local citizens and convert it into public use. In different countries, different terms are used: eminent domain, compulsory
purchase, land acquisition, etc. See Besley and Ghatak (2010) for summarized expropriations in selected countries throughout history, and see Anseeuw et al.
(2012) for large-scale land acquisitions, which cover 200 hectares or larger, in the developing countries during 2000–2012.
2
In the existing literature, the economic costs of state expropriation range from undermining investment incentives (e.g., Besley, 1995; Jacoby et al., 2002)
to resulting in long-term underdevelopment (e.g., Acemoglu et al., 2001, 2002).
3
That is, the state needs to extract revenue and resources from its citizens to provide public goods. For example, Tilly (1990, p. 96) argues that, as one of a
state’s essential minimal activities, extraction is crucial for the state to support other essential minimal activities; Besley and Ghatak (2010, p. 4560) point out
that it is common that expropriation supported by legislation was used as a public policy instrument to promote railway construction, milling, and mining during
1870–1910 in the United States (today’s paragon for upholding property rights); Besley and Persson (2011, p. 6) argue that the state plays an extractive role in
building and strengthening its fiscal ability.

https://doi.org/10.1016/j.jdeveco.2022.102985
Received 15 April 2021; Received in revised form 27 June 2022; Accepted 26 September 2022
Available online 3 October 2022
0304-3878/© 2022 Elsevier B.V. All rights reserved.
W. Sha Journal of Development Economics 160 (2023) 102985

Fig. 1. Share of households who experienced land expropriation. Notes: The red circles indicate the percentages of households who had their land expropriated in the survey
years: 2010, 2012, 2014, 2016, and 2018. In the first wave (2010), the survey asks about all the expropriation events of households in history (i.e., ever-expropriated), thus the share is
higher in this year. The sample includes the rural area only.
Source: China Family Panel Studies.

China provides an excellent empirical setting to study the political petitioned the provincial government but received no satisfactory reso-
consequences of state expropriation. As the world’s largest developing lution. After interacting with local government officials, some villagers
country, China has employed land expropriation to support develop- were arrested and jailed; and one died after being arrested by the
ment policies for several decades. Local governments have the authority police. During the process, the villagers gradually lost their trust in
to seize agricultural land and provide relatively low compensation. It the local government and called for an investigation by the central
is striking that, according to the China Family Panel Studies (CFPS), a government. Finally, a peaceful agreement between the villagers and
nationally representative longitudinal survey, the proportion of house- the government was reached after a top provincial leader intervened
holds who have had at least some of their land expropriated by 2018 and acknowledged the villagers’ basic demands.
was 30% (see Fig. 1). There is no doubt that land expropriation has To quantify the political costs of land expropriation, I focus mainly
contributed to China’s economic development and growth by facilitat- on two outcomes: individuals’ trust towards local government officials
ing infrastructure construction and urbanization.4 However, are there (‘‘political trust’’) and whether individuals experienced conflicts with
political costs associated with this development policy? And if so, is local government officials (‘‘political conflict’’). These two outcomes are
it possible for the government to mitigate the costs? Little is known important for several reasons. First, political trust is the foundation for
quantitatively about these questions, although they are of great interest political support and regime legitimacy (Newton et al., 2018). Second,
and policy relevance. both political trust and political conflict are critical to government
This paper argues that land expropriation incurs political costs for efficiency, especially the effectiveness of policy implementation. For
the Chinese government. The Wukan event, which attracted worldwide instance, Bargain and Aminjonov (2020) show that the compliance
attention, provides a vivid illustration.5 In September 2011, Wukan with public health policies during the COVID-19 pandemic in different
villagers launched large-scale violent protests against the government European countries depends on the level of political trust prior to
after they discovered that the village and township governments trans- the crisis. Last but not least, using these two measures, this study
ferred more than half of their agricultural land to a private real estate investigates how both political attitudes and behavior are shaped by
company without ex-ante negotiation or providing proper compensa- government expropriation. The political consequences of land expro-
tion to villagers. Some managers including the vice-CEO of the private priation identified in this paper also are informative for understanding
company were village cadres, suggesting that corruption may have the political situation in today’s China.
occurred at the local level.6 Before organizing the protests, the villagers I construct a national individual-level panel data set from the CFPS
with more than 8,000 households and 25,000 individuals for the period
2010–2018, which enables me to exploit rich variation in China’s land
4 expropriation. In the sample, I only include rural villages, i.e., those
Most of the expropriated land is used for building infrastructure
(e.g., roads and dams), establishing economic special zones and industrial that have a village committee (Chinese: Cunweihui). This is to ensure
parks, and urbanizing rural and suburban areas, etc. Theoretically, Xiong that the control group include only households and individuals that
(2018) constructs a growth model of the Chinese economy in which firm have agricultural land and thus are ‘‘eligible’’ for treatment.7 This paper
productivity is boosted by infrastructure investment provided by local govern- focuses on state expropriation of agricultural land of rural households
ments. Empirically, Lu et al. (2019) show that in China establishing special not programs such as housing demolition that mostly occur in urban
economic zones (SEZ) increases capital investment, employment, output, pro- areas and typically provide much more generous compensation. In the
ductivity, wages, and number of firms in the designated areas; Banerjee et al. data, land expropriation occurs at different periods for different house-
(2020) find that access to transportation networks moderately caused an
holds, I thus employ a generalized difference-in-differences approach
increase in per capita GDP level across sectors in China; He et al. (2020) find
that compares changes in individual’s political trust or political conflict
that China’s expressway system created faster growth in GDP for poor rural
counties.
before and after experiencing land expropriation with changes among
5
Wukan is a village located in Lufeng County, Guangdong Province (one individuals not experiencing expropriation during the same period. This
of the most developed provinces in China). The detail of the Wukan event is identifies the treatment effects of land expropriation on the treated.
covered by the Chinese and international media, such as South China Morning
Post and New York Times. Also, see He and Xue (2014) and Mattingly (2019)
7
for more detail about the Wukan event. Most households and individuals in urban villages, which have a commu-
6
The Wukan event also is known as an anti-corruption protest. See Chen nity committee (Chinese: Juweihui), have no agricultural land, and thus are
and Kung (2019) for corruption in China’s land market. never-takers of land expropriation.

2
W. Sha Journal of Development Economics 160 (2023) 102985

I begin by estimating the overall political impacts. The empirical about the expropriation projects at the village level, (ii) high govern-
finding is that if households have directly experienced land expropri- ment transparency level, and (iii) low government corruption level.
ation, both political beliefs and political behavior of individuals in Third, the political costs also can be reduced or even eliminated in
these households are affected. Specifically, exposure to land expropri- villages that experience infrastructure improvement (e.g., gaining ac-
ation decreases individuals’ trust towards local government officials cess to roads and railways or providing electricity or tap water) during
by 0.07 standard deviations, which is equal to about 36% of the the same period when villagers have their land expropriated. Last, the
mean difference in political trust between individuals with Chinese adverse political impacts are less salient for households with at least
Communist Party (CCP) membership and those without. In addition, one member working in the non-agricultural sector before having their
exposure to land expropriation increases the probability of having land expropriated. In other words, the political costs appear to arise
conflicts with local government officials by about 2.5%, compared mainly from perceived unfairness of compensation and nontransparent
to a sample mean of 5%, which means it increases the incidence of governance and can be reduced when projects generate public benefits
experiencing conflicts with government officials by around 50%. These or households depend less on agriculture.
adverse effects imply that expropriating land is indeed politically costly This study contributes first to the literature on the role of the
for local governments. The results are robust to a variety of internal state in promoting economic development. The common point in this
validity checks. literature is that the state has to be strong enough to provide se-
The empirical analysis also yields three important findings, which cure property rights and market institutions and constrained enough
further have some important political implications. First, the dynamic to minimize expropriation risk facing economic actors (e.g., North,
estimates show that the effects on both political outcomes are not per- 1981; North and Weingast, 1989; Acemoglu et al., 2005; Besley and
sistent, instead, they eventually attenuate to zero about four years after Ghatak, 2010; Besley and Persson, 2011; Acemoglu and Robinson,
treatment. Second, I do not find statistically significant spillover effects, 2012, 2019). However, Bardhan (2012, 2016) pointed out that ignoring
suggesting that the control group of non-expropriated individuals is the developmental role of the state may limit our understanding of the
not affected by the expropriation of the land of their neighbors. Last, diversity and complexity of the development process. The state also
individuals subject to expropriation also negatively assess government needs to play a critical, active role in ending poverty and in promoting
service and quality, but the effects are significant and remarkable only development more generally (see Page and Pande, 2018 for an excellent
in the short run. Taken together, these results suggest that the adverse review). By examining the political impacts of land expropriation which
political impacts are not persistent and restrictive to those affected is employed by the Chinese government to develop the economy, this
only. In other words, to some extent, land expropriation in rural areas study attempts to provide some new insight on the tension between
may not pose a fatal threat to China’s political regime. property rights and development goals. The findings suggest that in the
The Wukan example is by no means the only way in which how land context of state expropriation in developing countries, there could be a
takings occur in China. It epitomizes the worse possible features of land trade-off between property rights and potential development benefits
seizures (i.e., low compensation, no ex-ante negotiation, corruption, when (i) there is sound local governance and development projects
abuse of enforcement, no public benefits). Arguably, the outcomes are well implemented and (ii) the government uses the expropriated
could be different if the expropriation does not occur in that way. For resources to deliver public goods to citizens.
instance, one may ask what would be the consequences if the projects Second, this study complements the empirical literature on the
are well managed and justified by local governments and/or the land impacts of state expropriation (or extractive institutions more broadly).
is taken for public use (e.g., expressways and high-speed rail). To this Recently, Chen and Yeh (2020) show that state expropriation in the
end, I next turn to consider the role of compensation in determining the US increases infrastructure construction and growth and raises racial
political effects, and the extent to which the adverse political impacts inequality. Some scholars have also documented that expropriation
depend upon governance quality, the public benefits of projects, and risk reduces investment incentives for households (e.g., Besley, 1995;
the importance of the agricultural land for affected households. Jacoby et al., 2002), as well as for firms (e.g., Johnson et al., 2002;
Of course, many factors could intermediate the relationship between Cull and Xu, 2005). In contrast, the secure land property rights induce
land expropriation and political outcomes. I provide four insights into long-term investment (e.g., Goldstein et al., 2018), reduce inter-sectoral
this link. First, I provide evidence that households’ economic well- misallocation of labor (e.g., de Janvry et al., 2015), increase rental
being is not significantly affected by land expropriation, given that activity (e.g., Chari et al., 2022), and enhance agricultural produc-
(i) land expropriation does not create negative shocks to total house- tivity (e.g., Lin, 1992; Banerjee et al., 2002; Goldstein and Udry,
hold income,8 and (ii) saving asset (bank deposits and cash) of affected 2008; Chari et al., 2022). At the most aggregate level, using cross-
households increases, even though it is not persistent. However, com- country data, Acemoglu et al. (2001, 2002) have shown that extractive
institutions have led to long-term underdevelopment.10 Compared to
pared to the market value of the land, the compensation is very low,
the rich empirical evidence examining the economic impacts of state
indicative of an uneven redistribution of land rents between farmers
expropriation, however, the existing literature has provided little em-
and local government and property developers, which is likely to be
pirical evidence, if any, on the political consequences of government
perceived by farmers as unfair. This view is supported by a comparison
expropriation. This paper attempts to extend this strand of literature
with urban housing demolition which produces large, positive wealth
by studying the political impacts of China’s land expropriation.
shocks to affected households and has no impact on political outcomes.9
Third, this paper speaks to the studies on the origins of political
Second, I find that the adverse political impacts can largely be avoided
trust. Previous studies have emphasized the role of government per-
by sound local governance as reflected by (i) availability of information
formance in shaping citizens’ trust in the government (e.g., Dahlberg
and Linde, 2018).11 Some scholars have recently started investigating
8 this question in the context of China, focusing on the abolition of
I find that although land expropriation decreases agricultural income, it
does not create negative shocks to total household income. Household-level school fees (Lü, 2014), educational content (Cantoni et al., 2017),
analysis offers one possible explanation for this result: households who had historical experience of famine (Chen and Yang, 2019a), and media
their land expropriated are more likely to have at least one migrated member censorship (Chen and Yang, 2019b). This study links the decline of
as well as to earn wage income that may offset the decline in agricultural
income.
9 10
Note that the political impacts of housing demolition are not necessarily Hall and Jones (1999) also provide a related analysis.
11
positive if affected household think they deserve the generous compensation See van der Meer (2018) for a comprehensive review in the political
package as former housing owners. science literature.

3
W. Sha Journal of Development Economics 160 (2023) 102985

citizens’ trust in the government to a public policy that is intended to To generate revenue from the primary land market, local govern-
develop the economy but could be implemented unevenly at the local ments adopt a two-stage strategy. First, local governments completely
level. Closely related to this paper, Cui et al. (2015) also studied the im- control the land by expropriating it with relatively low compensation
pacts of land expropriation on political trust using cross-sectional data and convert the legal status of the land’s ownership from collective to
in China. However, the large-scale panel data enables me to credibly state-owned to meet the requirement that only state-owned land can be
identify causality and to study dynamics; moreover, I use a nationally legally exchanged. The second stage is to ‘‘privatize’’ the land use rights
representative sample and focus on a more recent time period; and to a third party (e.g., real estate developers and infrastructure construc-
for the first time I study the political and economic conditions that tion companies) in the primary land market. Local governments are sole
can mitigate the adverse political effects of land expropriation. Last, legal sellers and are able to charge a relatively high price. Therefore, by
this paper is also related to the literature on the relationship between expropriating land at a low compensation price and selling it at a high
property rights and land conflict (e.g., Alston et al., 1998; Hidalgo et al., price, this two-stage strategy creates a huge amount of extra-budgetary
2010). Using qualitative methods, Guo (2001) also draws a connection revenue for local governments.12
between land expropriation and rural conflicts in China during the Given some features embedded in Chinese institutions, land expro-
1990s. priation may have adverse political consequences. In terms of economic
The rest of this paper is organized as follows. Section 2 presents institutions, land property rights are poorly protected, with ownership
important background information. Section 3 introduces the data. Sec- being controlled by the state. The state monopolizes the authority of
tion 4 explains the empirical strategy. Section 5 reports the main results transferring the land to a third party, which bans the possibility of
on political impacts, including baseline estimates, robustness checks, households transferring the land use rights to companies themselves.
dynamics, spillovers, and effects on additional outcomes. Section 6 Consequently, local governments can offer relatively low compensation
provides evidence on the political and economic conditions under to affected households compared to the market value of the land.
which the adverse political impacts of expropriation can be mitigated. In terms of political institutions, local political leaders are usually
Section 7 concludes. accountable to upper-level leaders rather than citizens because of the
centralized personnel control system.13 The expropriation process thus
2. Land expropriation in China oftentimes involves aggressive official behavior. Because the media
is controlled or heavily censored by the government, citizens who
In rural China, a unique feature of the (agricultural) land property have their land expropriated and feel the compensation is unfair have
rights is that the land use rights are private but the ownership remains limited space to voice their concerns. Although these features make it
collectively owned. (The ownership of urban land is state-owned.) After possible for the Chinese government to employ land expropriation as
the Rural Land Contracting Law (RLCL) was carried out in 2003, rural an instrument to develop the economy, it is also extractive in the sense
households obtain well-defined and exclusive use rights by contracting that property rights protection is undermined, as noted by Acemoglu
the land from the village collective, generally with a period of 30 years. and Robinson (2012, 2019).
While the household contractor cannot be changed, the use rights Whether and how land expropriation has adverse political conse-
(i.e., carrying out production separately and claiming residual income quences depends mainly on four critical aspects. First, whether the
exclusively) can be legally transferred in the rental market. Although compensation is fair affects the political impacts of land expropriation.
some scholars believed that the RLCL has strengthened the land rights The Chinese central government sets up a basic scheme of compensa-
in rural China (Chari et al., 2022), the land tenure is still insecure and tion, which varies across localities. The compensation is not based on
often faces the risk of expropriation, which largely has to do with the the market value of the land itself, instead, local governments first es-
ownership structure. timate the average market value of the annual yield of crops cultivated
The Chinese laws (e.g., Constitution and Land Administration Law) on the land for the preceding three years and then offer compensation
allow local governments to expropriate land from local households who which is no more than 30 times of the estimated value.14 In practice,
are actually operating it. It is codified in the Constitution of the People’s as shown in Panel A of Fig. 2, the main form of compensation is
Republic of China that ‘‘[...] in accordance with the laws, the state can cash (about 71%). Other forms of compensation, such as allocating
acquire or expropriate citizens’ private properties for the need of the another piece of land, transferring hukou from rural to urban, providing
public interest after providing compensation.’’ Likewise, similar articles jobs, and providing pension insurance, etc., are also possible but are
regarding the land property are codified in the Land Administration Law not popular. Remarkably, more than 17% of households receive no
of the People’s Republic of China. Under this arrangement of ownership compensation.
structure, the state (the ruler) tends to maximize his own rents (North, Much of the expropriated land is not used for public purposes and
1981). Compared to western countries where land property rights are thus is sold to companies in the primary land market (see Table A.1).
completely private and well protected, China’s land property rights are In general, the land price in the primary market, which is received by
insecure. local governments from paid companies, is much higher than the price
Two Chinese institutional arrangements provide incentives for local of expropriated land, which is paid by local governments to households.
governments to expropriate land. The first is China’s fiscal revenue- It is clear that local governments and companies have shared the huge
sharing systems. There are two major sources of fiscal revenue for local rents generated from the land, while households who were originally
governments: budgetary and extra-budgetary revenue. The budgetary operating the land just get a tiny part. This large price difference is
(tax) revenue is shared with the central government who takes a great indicative of an extremely uneven redistribution of rents between the
proportion, whereas local governments have exclusive control rights government and households. Given that information on the market
over extra-budgetary revenue (Han and Kung, 2015). Consequently, price of land is publicly available, households are well aware of this
local governments allocate more effort to obtain extra-budgetary rev-
enue which consists mainly of land conversion income, generating 12
See Fig. A.1 in the Appendix for annual land conversion income obtained
by transferring the land use rights to a third party in the primary
by local governments.
land market. The second is the performance-based evaluation system 13
It is widely known that China’s institutions are economically decentralized
for promotion of local leaders (Li and Zhou, 2005; Xu, 2011). Both but politically centralized, which Xu (2011) calls ‘‘regionally decentralized
infrastructure construction and real estate development are driving authoritarian (RDA) regime.’’
forces for economic development and growth in China, incentivizing 14
See Article 47 of the Land Administration Law of the People’s Republic of
local leaders to engage in more land expropriation (Wang et al., 2020). China for more detail.

4
W. Sha Journal of Development Economics 160 (2023) 102985

Fig. 2. Compensation methods. Notes: The figure plots the share of different compensation methods. For land expropriation, affected households can receive compensation as
follows: money (cash), allocating another piece of land, providing jobs, providing pension insurance, transferring hukou from rural to urban, other forms, or no compensation.
Source: China Health and Retirement Longitudinal Study (CHARLS).

disparity, and so often are not satisfied with the compensation offered Instead, local governments can solely determine expropriation acreage,
by the government, which may lead to political costs. compensation, and resettlement, without prior negotiation with af-
The second is the justification for state expropriation, namely the fected households. Moreover, from 1998 onward, local governments do
public interest. In China, before 1998 the state restricted the public not have to suspend land expropriation when affected households have
interest to the construction of infrastructure and government and mil- disputes regarding compensation and resettlement.18
itary facilities.15 But afterwards, the public interest was expanded to Nevertheless, once the proposed expropriation project is approved
include private construction (e.g., urbanization). What this means is by the provincial government (or the State Council for larger projects),
that local governments can expropriate collective land for any project the county or prefectural government (depending on which level of
related to urbanization. For example, commercial housing construction government is in charge) is required to make public announcements
since 1998 can be categorized as ‘‘public interest’’. that provide detailed information pertaining to compensation and re-
One caveat of my data is that it does not contain information on settlement and so on.19 In practice, the village committee (Chinese:
the usage of expropriated land. Information regarding such usage is Cunweihui) also will often make announcements. Affected villagers can
described by Ma and Mu (2020), who analyze data from the China raise disputes to and/or demand a public hearing from the county or
Household Finance Survey (CHFS, see: https://chfs.swufe.edu.cn/), a prefectural government, who may feel obligated to increase compen-
nationally representative survey.16 According to Ma and Mu (2020), sation standards and resettlement conditions. Although this will not
during 2013–2015, many land expropriation projects are related to stop the implementation of the projects, it does make more informa-
the public interest: among the households surveyed by the CHFS, 48% tion available and hence makes the expropriation procedures more
report that their expropriated agricultural land is used for highway and transparent, resulting in better governance and accountability of local
railway construction, and 24% report the land is used for community governments. Thus, political impacts are expected to differ with the
infrastructure construction. At the same time, projects that are not availability of information on expropriation.
necessarily beneficial for local households also account for a large There is rich variation in the quality of governance across re-
proportion. Among the households surveyed, 26% and 20% report the gions.20 Wang et al. (2019), who study China’s market reform process,
land was used for housing development or by firms, respectively.17 calculate a marketization index at the provincial level for the 2008–
Given that public projects can improve infrastructure conditions and 2016 period, which reveals substantial differences across provinces. Nie
the living environment for local residents and thus can be justified et al. (2019), who study the government-business relationship in China,
in a easier way, it is expected that the political impacts may differ provide indices of government corruption and transparency at the city
depending on the usage of expropriated land. level in 2018, which also shows a huge dispersion across cities. Both
The third crucial aspect is procedural due process, which is related studies suggest that the quality of governance varies across regions in
to the quality of governance at the local level and may vary across China. I conjecture that local governments with a higher quality of
localities. Before 1998, the laws stipulate that local governments have governance handle land expropriation projects in a better way. Below,
to negotiate with affected households regarding expropriation acreage, I also examine the role of governance quality in mitigating the adverse
political impacts of land expropriation.
compensation, and resettlement. These relevant legal provisions were
Last but not least, to what extent the households depend on the
deleted from the 1998 amendment to the Land Administration Law.
land itself also could affect the political costs of land expropriation.
China has experienced a great economic transformation with rapid
15
Of the various legal stipulations pertaining to land expropriation, signifi- industrialization, urbanization, and rural-to-urban migration, which is
cant changes happened in 1998 when the state passed an amendment to the accompanied by numerous non-agricultural employment opportunities.
Land Administration Law. See Zhang and Feng (2015) for a detailed review of
legal changes regarding land expropriation in China since 1949.
16 18
See Table A.1 in the Appendix for more detail, which presents the relevant For example, see the Regulations on the Implementation of the Land Admin-
numbers that are reported in Table 1 and Footnote 28 of Ma and Mu (2020). istration Law of the People’s Republic of China (2014 Revision) for more detail
I cannot calculate the numbers myself because the CHFS does not make the about these two features of land expropriation in China.
19
data contain information on the usage of expropriated land publicly available See the Measures for Announcement of Land Expropriation (2010
to outside users. Amendment) for more detail.
17 20
Note that the percentages do not add up to one as one household can See Bardhan (2020) for a general discussion about China’s governance
report multiple usages. system.

5
W. Sha Journal of Development Economics 160 (2023) 102985

Table 1
Summary statistics for key variables.
Source: China Family Panel Studies.
Pooled years 2010 2012 2014 2016 2018
Obs. Mean Std. dev. Mean Mean Mean Mean Mean
(1) (2) (3) (4) (5) (6) (7) (8)
Land expropriation 148618 0.1966 0.3974 0.1290 0.1650 0.2070 0.2291 0.2645
Trust towards local officials 97557 5.2024 2.6603 – 5.1051 5.2838 5.1389 5.2861
Conflict with local officials 86785 0.0503 0.2186 0.0837 0.0355 0.0482 0.0419 –

Notes: Unit of observation is the individual-year. The table reports descriptive statistics of key explanatory variable and political outcomes for
all years, and 2010, 2012, 2014, 2016, and 2018. Land expropriation is a dummy that is 1 if the individual is in a household whose land has
ever been expropriated. Trust towards local officials is a categorical variable ranged from 0 (extremely low trust) to 10 (extremely high trust).
The 2010 wave does not contain information on political trust. Conflict with local officials is a dummy that equals to 1 if the individual had
conflicts with local government officials in the past year for the 2012, 2014, and 2016 waves; in the 2010 wave, conflict with local officials is
a dummy that is 1 if the individual has ever had conflicts with local officials, therefore, the corresponding ratio in 2010 is much higher. The
2018 wave does not contain information on political conflict.

Hypothetically, if local households have members working in the non- (15) or above, so the overall sample size for political trust is bigger
agricultural sector and thus rely less (or even little) on the agricultural than that for political conflict. Finally, I construct a five-year panel with
land to generate food and income, then it may matter little when the more than 8,000 households and 25,000 individuals.
land is expropriated, whereas those with all adult members working in The empirical analysis largely uses the following three variables: (i)
agriculture should politically react more to land expropriation. In this exposure to land expropriation, the treatment indicator which varies
regard, I expect the political impacts to be different for households with at the household level. I assign one to the individual if some of his
and without members working in the non-agricultural sector before or her family’s land was expropriated anytime in the past and zero
experiencing land expropriation. otherwise; (ii) trust towards local government officials, which varies
at the individual level and is a categorical variable ranging from zero
3. Data (extremely low trust) to ten (extremely high trust); and (iii) conflict
with local government officials, which also varies at the individual
This paper uses various data sets. I rely mainly on the China Family level. In the 2012, 2014, and 2016 waves, the conflict variable is a
Panel Studies (CFPS, see http://www.isss.pku.edu.cn/cfps/), which is a dummy that is equal to one if the individual has conflicts with local
nationally representative longitudinal survey of Chinese communities, government officials in the past 12 months and zero otherwise. In the
families, and individuals. It was launched in 2010 and has been carried 2010 wave, it takes the value of one if the individual has ever had
out every two years since then.21 My main sample and key variables, conflicts with local officials and zero otherwise.
Table 1 reports descriptive statistics for the treatment indicator
which I will describe in detail below, are constructed from the CFPS.
and political outcomes for all years, and in 2010, 2012, 2014, 2016,
I also make use of the 2014 wave of the China Health and Retirement
and 2018. On average, 20% of individuals in the sample constructed
Longitudinal Study (CHARLS, see http://charls.pku.edu.cn/), which is
from the CFPS rural sample have experienced land expropriation. In
also a nationwide and representative survey, to calculate the expropri-
total, about 26% of individuals have had at least some of their land
ated price of agricultural land and the share of different compensation
expropriated (column 8).22 The average level of trust towards local
methods for land expropriation.
government officials of individuals is 5 on a scale of 0–10. Nearly 5%
My main sample is constructed from the CFPS in the following
of individuals have had conflicts with local government officials during
way. First and foremost, although the CFPS covers both rural and
2010–2016. This seems high, possibly due to the broad definition of
urban areas, I restrict the sample to rural villages in which all house-
political conflict used in the CFPS. It is worth emphasizing that I
holds cultivate agricultural land and hence are potential victims of
exclude always-treated units in the regressions, so each regression is
land expropriation. Rural villages have a village committee (Chinese:
run by using a data set with first period excluding the treated units.
Cunweihui) as their administrative body, while urban villages have
For example, the trust (conflict) regressions exclude those who have
a community committee (Chinese: Juweihui). This condition, which
their land expropriated in or before 2012 (2010).
drops about 30% of the villages sampled by the CFPS, ensures that In Fig. 3, Panels A and B plot the distribution of political trust and
the control group is less heterogeneous and more comparable because political conflict by year across groups, respectively. In Panel A, one
most households and individuals in these excluded urban villages have can see that the percentage of individuals who report their political
no agricultural land and hence cannot have their land expropriated. trust level under 5 is higher in the expropriated group than that in the
In the Appendix, Fig. A.2 plots the time trend of political outcomes non-expropriated group. This pattern holds consistently for all years.
by group, indicating that most urban individuals who are not subject Similarly, one can also see from Panel B that the expropriated group has
to land expropriation have a differential time trend of political trust. a higher proportion of individuals who have experienced conflicts with
Hence, including this group in the sample may lead the control group local government officials than the non-expropriated group in all years.
to be inappropriate, as it violates the parallel trends assumption of Apparently, individuals who have had some of their land expropriated
the difference-in-differences estimator (see more detailed discussion in tend to report a lower level of trust in the government and to be more
Section 4). likely to experience conflicts with local officials. That is to say, there is
Second, up to now, five waves of data (2010, 2012, 2014, 2016, a significant association between land expropriation and the political
and 2018) are publicly available. However, the 2010 (2018) wave does outcomes of interest.
not contain information on political trust (political conflict), so the My key variables suffer from some concerns because of the CFPS sur-
time period spans 2012, 2014, 2016, and 2018 for political trust and vey design and possible political sensitivity of answering such questions
2010, 2012, 2014, and 2016 for political conflict. Third, information in China. Moreover, the sample also suffers from attrition issue. Here I
on political trust (political conflict) is available for individuals aged 10 discuss these concerns and how they could lead to biased estimates.

21 22
A data-maintaining wave is done in 2011, but it is not a regular wave of This is slightly less than 30% which is reported in Fig. 1 because Fig. 1
the survey. counts multiple times of land expropriation of a household separately.

6
W. Sha Journal of Development Economics 160 (2023) 102985

Fig. 3. Distribution of political trust and political conflict. Notes: Panels A and B plot the distributions of political trust and political conflict by group in different years, respectively.
Political trust is a categorical variable ranged from 0 (extremely low trust) to 10 (extremely high trust). Political conflict is a dummy that equals to 1 if the individual had conflicts
with government officials in the past year for the 2012, 2014, and 2016 waves and 0 otherwise; in the 2010 wave, it is a dummy that is 1 if the individual has ever had conflicts
with local officials.
Source: China Family Panel Studies.

One problem of the CFPS is that the survey is conducted every two One may also have concerns about attrition. If individuals whose
years, but except for the 2010 wave, the questions utilized by this land was expropriated are more likely to drop out of the sample, my
study only ask about what happened in the past 12 months, resulting estimates could be biased. To address this concern, I plot the attrition
in an incomplete measurement of expropriation events between con- patterns for the full, expropriated, and non-expropriated samples in
secutive survey waves. I thus cannot rule out the possibility that some the Appendix, Fig. A.3, revealing no evidence that individuals subject
households experienced expropriation during the sample period but are to land expropriation are any more likely to attrit from the sample.
recorded as not having their land expropriated. Given this measurement Additionally, Table A.3 presents the result of a simple regression-based
error, the estimates should be viewed as a lower bound. Another test, indicating that exposure to land expropriation is not significantly
measurement error could arise if among individuals subject to land associated with moving out of the sample and the magnitude of the
expropriation only those who become frustrated with the government
coefficient is negligible.
are more willing to report their experience of expropriation, leading
to a mechanical negative relationship between land expropriation and
4. Empirical strategy
political trust. This is less of a concern since I do not find evidence of a
statistically significant correlation between land expropriation and life
satisfaction or depression (see Table A.2). To formally study the political consequences of land expropriation,
Another problem of the CFPS is that it does not provide a clear I estimate the following generalized difference-in-differences model,
definition for political conflict (Chinese: Chongtu), meaning that conflict exploiting the variation in the timing of exposure to land expropriation
can take any form based on respondents’ own definition. Nevertheless, across households:
one can think of the minimal case as having disagreements with local
𝑃 𝑜𝑙𝑖𝑡𝑖𝑐𝑎𝑙𝑂𝑢𝑡𝑐𝑜𝑚𝑒𝑖ℎ𝑣𝑡 = 𝛽 × 𝐿𝑎𝑛𝑑𝐸𝑥𝑝𝑟𝑜𝑝𝑟𝑖𝑎𝑡𝑖𝑜𝑛ℎ𝑣𝑡 + 𝛼𝑖 + 𝜆𝑡 + 𝜖𝑖ℎ𝑣𝑡 , (1)
officials, the maximal case being participation in protests or physical
fights with officials. Given this situation, one should be cautious in where 𝑖 indexes individuals, ℎ households, 𝑣 villages, 𝑡 time periods.
generalizing the results to other contexts. 𝑃 𝑜𝑙𝑖𝑡𝑖𝑐𝑎𝑙𝑂𝑢𝑡𝑐𝑜𝑚𝑒𝑖ℎ𝑣𝑡 is either trust towards local officials or conflict
Given the political environment in China, one may worry that the with local officials. 𝐿𝑎𝑛𝑑𝐸𝑥𝑝𝑟𝑜𝑝𝑟𝑖𝑎𝑡𝑖𝑜𝑛ℎ𝑣𝑡 is the treatment indicator of
self-reported level of the two political outcomes contains significant interest, which is equal to 1 after the household has ever been expro-
reporting bias, especially for political trust. If the misreported level is priated and 0 otherwise. Note that the treatment status for individuals
similar or the same for an individual in all years, then this individual- within a household is the same. 𝛼𝑖 and 𝜆𝑡 are individual and time
specific, systematical reporting bias will be differenced out by including fixed effects, respectively. The error term 𝜖𝑖ℎ𝑣𝑡 is clustered at the village
individual fixed effects. Another possibility is that respondents over-
level, allowing for correlation across individuals within a village. 𝛽 is
report (under-report) the level of political trust (political conflict) only
the coefficient of interest that captures the treatment effect of state
after they have been expropriated, possibly because they now interact
expropriation on political trust or political conflict.
with the government frequently and fear being mistreated if they
Identification comes from how changes in the outcomes of interest
truthfully report. However, as depicted in Fig. 3, this is not likely to
be the case given that the expropriated group reports a lower level of are correlated with changes in exposure to land expropriation. In
political trust and a higher level of political conflict each year compared my data, land expropriation changes at different periods for different
to the non-expropriated group.23 Also, even if this were the case, the households. This enables me to control for both individual and time
empirical estimates could be viewed as lower bound estimates. fixed effects with identification based on within-individual before-after

23 may face much less pressure to self-censor trust towards local government
Additionally, Chen and Yang (2019a), who also utilize the CFPS to study
political trust, show that it might be the case in reality that the Chinese citizens officials.

7
W. Sha Journal of Development Economics 160 (2023) 102985

Table 2
Balance checks of time-invariant characteristics.
Source: China Family Panel Studies.
Expropriation
All Yes No Unconditional Conditional
Mean Std. dev. Mean Mean Diff. p-value Diff. p-value
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A: Individual-level Characteristics
Male 0.5012 0.5000 0.4977 0.5025 −0.0048 0.4208 −0.0016 0.6811
Han 0.8908 0.3119 0.8631 0.9009 −0.0379 0.0000 −0.0018 0.6499
CCP membership 0.1246 0.3302 0.1140 0.1284 −0.0144 0.0004 −0.0024 0.6395
Panel B: Household-level Characteristics
Family genealogy 0.2388 0.4264 0.2353 0.2400 −0.0048 0.6325 −0.0067 0.5123
Distance to nearest high school (km) 19.9515 27.3072 17.8725 20.6811 −2.8085 0.0000 0.7234 0.1826
Distance to nearest medical clinic (km) 1.9325 3.1307 1.7398 1.9996 −0.2598 0.0003 −0.0393 0.5949
Distance to nearest marketplace (min) 34.0081 47.1849 32.4839 34.5406 −2.0568 0.0587 −0.7262 0.5499
Panel C: Village-level Characteristics
Ancestral hall 0.1291 0.3357 0.1330 0.0833 0.0497 0.3946 −0.0586 0.3540
Any clan with population share >= 10% 0.7790 0.4154 0.7767 0.8056 −0.0288 0.6898 0.0126 0.8930
# of clans with population share >= 10% 1.8206 1.6190 1.8361 1.6389 0.1972 0.4836 0.1874 0.5521
Minority area 0.5142 1.3525 0.4846 0.8611 −0.3766 0.1090 −0.2594 0.4196
Natural resource area 0.5208 1.4158 0.4917 0.8611 −0.3694 0.1331 −0.3038 0.3342
Distance to town center (km) 5.5915 12.9028 5.1041 11.2917 −6.1877 0.0056 −4.4888 0.2765
Distance to county center (km) 27.5405 22.8210 26.7352 36.9583 −10.2232 0.0097 −7.1950 0.0445

Notes: Columns 1–2 present summary statistics; columns 3–4 show the means between expropriated and non-expropriated groups; columns 6–7 report
unconditional differences in means and the corresponding p-values; columns 8–9 report differences conditional on village/county fixed effects: that is,
I regress the characteristic on the land expropriation dummy each time, controlling for village fixed effects for Panels A–B and county fixed effects for
Panel C. Standard errors clustered at the village level. At the individual level, I check for gender dummy, han indicator, and Chinese Communist Party
(CCP) membership dummy; at the household level, I check for family genealogy dummy, the distance between the household’s geographical location
to the nearest high school (kilometers), medical clinic (kilometers), and marketplace (minutes by walking); at the village level. I check for whether
the village has a ancestral hall or a clan with population share >= 10%, the number of clans with population share >= 10%, whether it belongs to a
minority area or a natural resource area, and its distance to town center (kilometers) or county center (kilometers).

comparison. The individual fixed effects, 𝛼𝑖 , account for all unobserv- Chinese context given that land is only allocated to households in the
able individual-specific confounders that are associated with exposure villages where they are officially registered as residents. Furthermore,
to land expropriation. Thus, identification is driven by individuals who during the sample period there was no land reallocation in rural villages
experience changes in exposure during the sample period, who are in China (Zhao, 2020). Therefore, to a large extent, land expropriation
compared to those who do not experience a change in exposure. In cannot be manipulated by households and individuals.
this regard, this paper mainly estimates the treatment effects on the I now examine who is more likely to have their land expropriated.
treated. I also include time period fixed effects, 𝜆𝑡 , which absorbs In Table 2, I present the results of mean comparisons of time-invariant
unobservable temporal common shocks to all individuals. Note that characteristics between individuals (Panel A), households (Panel B),
in the regressions I exclude always-treated units. After controlling for and villages (Panel C) with and without land expropriation.24 Re-
these two sets of fixed effects, the identifying assumption is that any stricting attention to unconditional differences that are statistically
time-varying individual characteristics that affect the outcomes are significantly different from zero, I find that expropriated individuals
uncorrelated with the timing of land expropriation. are more likely to be from an ethnic minority and less likely to be CCP
My identification strategy creates two concerns. First, after con- members, indicating that individuals who are politically less powerful
trolling for both individual and time fixed effects, the main threat to are more likely to be expropriated. I also find that households who are
identification is the correlation between the timing of land expropria- closer to high schools (generally located in the county center in rural
tion and the evolution of political outcomes over time. Second, if the Chinese counties), medical clinics, and marketplaces are more likely to
control group of non-expropriated individuals is affected by the expro- have their land expropriated. It makes sense that land expropriation
priation of the land of their neighbors, the baseline estimates would projects are more likely to occur in relatively developed areas where
be biased towards zero. In what follows, I discuss several strategies to the value of the land is greater. Once I compare mean characteristics
address these concerns, validating the identifying assumption. of individuals and households within villages, however, there are no
It is worth emphasizing first that households (and individuals) longer any significant differences. At the village level, villages with a
shorter distance to town or county center are more likely to experience
do not have the power to influence land expropriation programs. In
land expropriation; the latter is still true even after I control for county
reality, land expropriation generally involves large-scale development
fixed effects. This again suggests that the government develops land
projects, which usually include constructing roads and dams, establish-
closer to urban centers. I do not find evidence that kinship group
ing economic development zones and industrial parks, urbanizing rural
influence land expropriation when I compare characteristics such as
and suburban areas, and so on. Thus, it is actually the county- and
whether a household has family genealogy or whether a village has any
prefecture-level governments that generally design and initiate these
ancestral halls or a clan whose population share is larger than 10%.25
projects, which require further approval by the provincial government
Taken together, these results suggest that the government selecting
or the State Council. This implies that households and individuals are
locations based on project needs rather than targeting a particular type
not typical participants in the decision-making process. An extreme
of households when carrying out land expropriation projects. Even
but illustrative example is the Three Gorges Dam in which none of
the affected households and individuals could influence government
policies. One may worry that even if households cannot influence 24
In Table A.4, I conduct a regression-based analysis and find similar results.
official plans, it is still possible that the plans will be anticipated 25
In rural China, the mean of population share of the second-largest clan
by households who in turn may behave strategically, for instance by is about 10% and the first two largest clans are most powerful (Xu and Yao,
moving to a new location. However, this is not a serious concern in the 2015). My results here are in contrast to the findings of Mattingly (2016).

8
W. Sha Journal of Development Economics 160 (2023) 102985

Table 3
The time-varying determinants of land expropriation.
Source: China Family Panel Studies.
Land expropriation
Household-year panel Individual-year panel
(1) (2) (3) (4) (5) (6) (7) (8)
Log HH income per capita (t-1) 0.0001 0.0002 −0.0014
(0.0018) (0.0020) (0.0030)
Log HH consumption per capita (t-1) −0.0006 −0.0023 −0.0052
(0.0031) (0.0033) (0.0059)
Log HH asset value per capita (t-1) 0.0020 0.0022 0.0051
(0.0020) (0.0022) (0.0039)
Trust towards local officials (t-1) 0.0011 0.0009 0.0009
(0.0009) (0.0009) (0.0011)
Conflict with local officials (t-1) −0.0032 −0.0106 −0.0128
(0.0076) (0.0106) (0.0123)
Dep. var. mean 0.0608 0.0604 0.0613 0.0609 0.0644 0.0641 0.0646 0.0646
Dep. var. SD 0.2389 0.2383 0.2398 0.2391 0.2454 0.2450 0.2458 0.2459
# of villages 1830 1846 1810 1721 1794 1836 1786 1516
# of observations 31537 30323 31297 28475 55625 65992 51994 44568
Adj. R-squared 0.1227 0.1228 0.1216 0.1212 0.1571 0.1439 0.1577 0.1464
Household fixed effects Yes Yes Yes Yes No No No No
Individual fixed effects – – – – Yes Yes Yes Yes
Time fixed effects Yes Yes Yes Yes Yes Yes Yes Yes

Notes: Unit of observation is the household-year for columns 1–4 and the individual-year for columns 5–8. The sample is a five-year panel
(2010, 2012, 2014, 2016, and 2018) for columns 1–4, a four-year panel (2012, 2014, 2016, and 2018) for column 5, a four-year panel (2010,
2012, 2014, and 2016) for column 6, and a three-year panel (2012, 2014, and 2016) for columns 7–8. Land expropriation is a dummy that
is 1 if the individual is in a household whose land has been expropriated in the past year. Household income, consumption, and asset value
per capita refer to average household net income, consumption, and asset value, respectively. Trust towards local government officials is a
categorical variable ranged from 0 (extremely low trust) to 10 (extremely high trust) in a given year. Conflict with local government officials is
a dummy that equals to 1 if the individual had conflicts with local government officials in the past year for the 2012, 2014, and 2016 waves,
and equals to 1 if the individual has ever had conflicts with officials in history for the 2010 wave. All household-level explanatory variables are
logged and all explanatory variables are lagged. Note that taking lag of variables will drop one period of data. Always-expropriated households
and individuals are excluded in the regressions. Robust standard errors in parentheses are clustered at the village (of current residence) level.
∗∗∗
: significant at 1%; ∗∗ : significant at 5%; ∗ : significant at 10%.

if there are still differences between treatment and control groups, to all households and individuals in a village. One caveat of including
including individual/household fixed effects will eliminate all threats village-by-year fixed effects is that many villages carried out only one
to identification that arise from differences in time-invariant character- project in a given year, therefore, accounting for village-by-year shocks
istics. Additionally, and perhaps more importantly, while the results of reduces the variation for identification. Second, to make time effects
balance checks suggest that individuals in different villages may sys- less restrictive at the village level, I further investigate how including
tematically differ from one another, the results are robust to including both time fixed effects and village-specific linear time trends affects
village-by-year fixed effects, a specification that compares individual- the results. Last, I account for time shocks at the county level by
level outcomes across the treatment and control groups within the same including county-specific time fixed effects, which is less restrictive
village and the same year. These treatment and control groups exhibit than village-year fixed effects.
no significant differences in characteristics. Moreover, I also interact In addition, to alleviate concern that the results could be driven
time-invariant characteristics with time dummies to further account for by time-varying factors, I include two time-varying covariates (age
differential time trends. dummies and educational-level dummies) in the regressions.
Turning to time-varying characteristics, I run panel regressions of In my research setting, a standard way to validate the identification
the treatment dummy on possible time-varying determinants, control- strategy is to check whether the treatment and control groups follow
ling for household/individual and time fixed effects. Table 3 reports the similar pretreatment parallel trends using an event-study design. To this
results. From columns 1–4, one can see that current household income, end, I estimate the following equation:
consumption, and asset value are not predictors of being expropriated
∑
2
in the next period, suggesting that it is not the case that poor house- 𝑃 𝑜𝑙𝑖𝑡𝑖𝑐𝑎𝑙𝑂𝑢𝑡𝑐𝑜𝑚𝑒𝑖ℎ𝑣𝑡 = 𝛽𝜏 × 𝟏(𝐸𝑥𝑝𝑟𝑜𝑝𝑟𝑖𝑎𝑡𝑒𝑑𝑃 𝑒𝑟𝑖𝑜𝑑ℎ𝑣𝑡 = 𝜏)
holds are more likely to be expropriated. From columns 5–7, we see 𝜏=−3,𝜏≠−1
that lagged political outcomes also are not significantly correlated with
+ 𝛼𝑖 + 𝜆𝑡 + 𝜖𝑖ℎ𝑣𝑡 , (2)
being treated, suggesting that the identification does not suffer from
reverse causality and that it is less likely to be the case that those who where 𝑖 indexes individuals, ℎ households, 𝑣 villages, 𝑡 time periods.
have a lower level of trust in the government or a higher likelihood 𝑃 𝑜𝑙𝑖𝑡𝑖𝑐𝑎𝑙𝑂𝑢𝑡𝑐𝑜𝑚𝑒𝑖ℎ𝑡 is either political trust or political conflict. 𝜏 = −3,
of having conflicts with local officials are more likely to experience −2, 0, 1, 2 (the gap between every two consecutive periods is two
land expropriation. Column 8 presents the results of a regression in- years); Period −1 (𝜏 = −1, base period) is used as the reference group
cluding both household- and individual-level covariates. Overall, being and hence omitted in the regressions; Period 0 (𝜏 = 0) is the period
expropriated is not strongly associated with many characteristics after right after the expropriation occurred. 𝐸𝑥𝑝𝑟𝑜𝑝𝑟𝑖𝑎𝑡𝑒𝑑𝑃 𝑒𝑟𝑖𝑜𝑑ℎ𝑣𝑡 stands
including all fixed effects, which supports the randomness of the timing for the period relative to the base period for household ℎ in year 𝑡.
of land expropriation. A set of dummies, 𝟏(𝐸𝑥𝑝𝑟𝑜𝑝𝑟𝑖𝑎𝑡𝑒𝑑𝑃 𝑒𝑟𝑖𝑜𝑑ℎ𝑣𝑡 = 𝜏), indicates whether
One major concern for difference-in-differences estimates is differ- in year 𝑡 household ℎ is in the 𝜏th period before or after the base
ential time trends, especially given the long period of time covered period. The error term 𝜖𝑖ℎ𝑣𝑡 is clustered at the village level. In this
by this study. To address this concern, instead of including time fixed specification, the 𝛽𝜏 ’s are informative, we can check for (i) the parallel
effects, I restrict the variation in the following three ways. First, I allow pre-trends assumption when 𝜏 = −3, −2; (ii) the immediate effects of
time effects to be village-specific, absorbing shocks that are common land expropriation when 𝜏 = 0; and (iii) the persistence of the effects

9
W. Sha Journal of Development Economics 160 (2023) 102985

Table 4
Effects of land expropriation on political outcomes.
Source: China Family Panel Studies.
Trust towards local government officials
Exclude Balanced
movers panel
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A: Effects on political trust
Land expropriation −0.1951∗∗ −0.1387∗∗ −0.1687∗∗ −0.1669∗∗ −0.2065∗∗ −0.2288∗∗ −0.2189∗∗ −0.2238∗∗
(0.0819) (0.0644) (0.0845) (0.0786) (0.0818) (0.0980) (0.0902) (0.0951)
Land expro. mean 0.0688 0.0688 0.0688 0.0688 0.0688 0.0699 0.0675 0.0721
Land expro. SD 0.2531 0.2531 0.2531 0.2531 0.2532 0.2550 0.2509 0.2587
Dep. var. mean 5.2548 5.2548 5.2548 5.2548 5.2524 5.2618 5.2788 5.3640
Dep. var. SD 2.6610 2.6610 2.6610 2.6610 2.6595 2.6578 2.6619 2.6681
# of villages 2043 2043 2043 2043 2039 450 451 1116
# of observations 78336 78336 78336 78336 77693 61991 70937 44519
Adj. R-squared 0.3238 0.3392 0.3293 0.3313 0.3267 0.3277 0.3211 0.3292
Conflict with local government officials
Exclude Balanced
movers panel
(1) (2) (3) (4) (5) (6) (7) (8)
Panel B: Effects on political conflict
Land expropriation 0.0253∗∗ 0.0129∗ 0.0198∗∗ 0.0177∗ 0.0255∗∗ 0.0230∗∗ 0.0242∗∗ 0.0260∗∗∗
(0.0102) (0.0077) (0.0099) (0.0097) (0.0102) (0.0105) (0.0108) (0.0099)
Land expro. mean 0.0734 0.0734 0.0734 0.0734 0.0729 0.0735 0.0725 0.0735
Land expro. SD 0.2608 0.2608 0.2608 0.2608 0.2600 0.2609 0.2593 0.2609
Dep. var. mean 0.0481 0.0481 0.0481 0.0481 0.0481 0.0490 0.0486 0.0492
Dep. var. SD 0.2140 0.2140 0.2140 0.2140 0.2140 0.2158 0.2151 0.2163
# of villages 1502 1502 1502 1502 1497 454 454 841
# of observations 73459 73459 73459 73459 72789 65220 69748 40673
Adj. R-squared 0.1404 0.1739 0.1521 0.1503 0.1396 0.1439 0.1441 0.1498
Individual fixed effects Yes Yes Yes Yes Yes Yes Yes Yes
Time fixed effects Yes No Yes No Yes Yes Yes Yes
Village FEs X time FEs No Yes No No No No No No
Village FEs X time trend No No Yes No No No No No
County FEs X time FEs No No No Yes No No No No
Time-varying controls No No No No Yes No No No
Time-invariant controls X time FEs No No No No No Yes No No

Notes: Unit of observation is the individual-year. The sample is a four-year panel (2012, 2014, 2016, and 2018) for Panel A, and a four-year panel
(2010, 2012, 2014, and 2016) for Panel B. Columns 1–6 use the main sample; column 7 excludes those who move out of the villages where they reside
in when being surveyed at the first time; and column 8 uses a balanced panel. Individuals who have their land expropriated in or before 2012 (2010)
are excluded in the regressions for Panel A (B). Trust towards local government officials is a categorical variable ranged from 0 (extremely low trust)
to 10 (extremely high trust) in a given year. Conflict with local government officials is a dummy that equals to 1 if the individual had conflicts with
local government officials in the past year for the 2012, 2014, and 2016 waves, and equals to 1 if the individual has ever had conflicts with officials
in history for the 2010 wave. Land expropriation is a dummy that is 1 if the individual is in a household whose land has ever been expropriated in
a given year. Time-varying control variables include age dummies and educational-level dummies. Time-invariant control variables include all of the
variables in Table 2. Robust standard errors in parentheses are clustered at the village (of current residence) level. ∗∗∗ : significant at 1%; ∗∗ : significant
at 5%; ∗ : significant at 10%.

when 𝜏 = 1, 2, 3. Note that (i) I drop always-treated individuals in the direction. In reality, whether one has her or his family’s land expropri-
regressions; and (ii) the fact that the sample includes a large proportion ated and whether one is in a village that carries out land expropriation
of never-treated individuals (‘‘a pure control group’’) could mitigate programs are likely highly correlated.26 Thus, it may not be proper
the underidentification concern in event study designs in absence of to examine the spillover effects by including individual- and village-
such a control group (Borusyak et al., 2022). Furthermore, I also report level land expropriation experience in the same regression. To this end,
the results obtained using the method proposed by Sun and Abraham I use two other ways to study the spillover effects. First, I augment
(2021), addressing the concern that the estimates of 𝛽𝜏 ’s could be Eq. (1) with the share of individuals subject to land expropriation
contaminated and treatment effect heterogeneity may lead to apparent within counties to examine the possible spillovers across individuals
pre-trends since different households receive the treatment in different within counties. Second, I exclude treated units and construct a sample
periods (i.e., there is variation in treatment timing across units). that includes the control individuals only. Then, I regress the political
Up to now, I have not considered the spillover effects of land outcomes of interest on the ratio of treated villagers by using the newly
expropriation projects from individuals whose land was expropriated constructed sample to examine the possible within-village spillovers.
to those whose land was not expropriated. Such spillovers could exist,
especially across individuals within regions (e.g., villages or counties). 5. Political impacts
In the empirical setting of this paper, I do not have a strong prior on the
direction of the spillovers. For instance, unaffected individuals might 5.1. Baseline results and robustness checks
have sympathy for neighbors whose land is expropriated and hence
distrust the government and have conflicts with local officials. On the The baseline results, obtained from estimating Eq. (1), suggest
other hand, since individuals not experiencing land expropriation could that land expropriation has adverse political consequences (Table 4).
be better-off if land expropriation helps improve the infrastructure
conditions or increase the provision of public goods, there might be
26
no spillover effects, or the spillovers may even work in the opposite I thank both referees for pointing this out.

10
W. Sha Journal of Development Economics 160 (2023) 102985

Table 5
Effects of land expropriation on non-political trust.
Source: China Family Panel Studies.
General Trust in Trust in Trust in Trust in Trust in
trust parents neighbors Americans strangers doctors
(1) (2) (3) (4) (5) (6)
Land expropriation 0.0594 −0.0471 −0.1156 −0.0810 0.0250 −0.1105
(0.0637) (0.0474) (0.0730) (0.0877) (0.0759) (0.0749)
Land expro. mean 0.0688 0.0688 0.0688 0.0693 0.0689 0.0687
Land expro. SD 0.2531 0.2530 0.2531 0.2539 0.2532 0.2530
Dep. var. mean 2.8310 9.2460 6.6619 2.3758 2.0394 6.9547
Dep. var. SD 1.9929 1.5369 2.2138 2.5087 2.1420 2.3375
# of villages 2045 2044 2047 2035 2045 2045
# of observations 78515 78373 78583 76689 78381 78557
Adj. R-squared 0.2307 0.2159 0.3063 0.2925 0.2713 0.2824
Individual fixed effects Yes Yes Yes Yes Yes Yes
Time fixed effects Yes Yes Yes Yes Yes Yes

Notes: Unit of observation is the individual-year. The sample is a four-year panel (2012, 2014, 2016, and 2018). Individuals who have their
land expropriated in or before 2012 are excluded in the regressions. General trust takes value of 1 (low trust) or 5 (high trust). Trust towards
parents, neighbors, doctors, strangers, and Americans are categorical variables ranged from 0 (extremely low trust) to 10 (extremely high trust).
Land expropriation is a dummy that is 1 if the individual is in a household whose land has ever been expropriated in a given year. Robust
standard errors in parentheses are clustered at the village (of current residence) level. ∗∗∗ : significant at 1%; ∗∗ : significant at 5%; ∗ : significant
at 10%.

Individuals who have their land expropriated have trust towards local In summary, the estimates reported in columns 2–6 of Table 4
government officials that is lower by 0.2, which is equivalent to about show that the baseline results are robust to using various ways of
0.07 standard deviations (Panel A, column 1). The sample mean of controlling for differential time trends, the main identification threat to
political trust for individuals with CCP membership is 5.69 and is the empirical strategy of this paper, confirming that land expropriation
5.15 for non-party members; thus, this effect equals about 36% of the leads to adverse political impacts.
mean difference in political trust between individuals with and without Column 7 of Table 4 excludes movers (including migrants) who
CCP membership. In Panel B, column 1 shows that having one’s land move out of villages where they are officially registered as residents,
expropriated also increases the probability of having conflicts with local addressing potential endogeneity caused by migration. For example,
government officials by 2.5%, compared to a sample mean of 5%, movers or migrants may experience differential changes in political
which means it increases the incidence of having conflicts with local outcomes after the land expropriation projects due to reasons other
officials by about 50%. This large effect is consistent with the high than land expropriation, which in turn drives the baseline results.
frequency of land-related conflicts in China.27 The baseline findings are Column 8 of Table 4 reports the results using a balanced panel, which
not sensitive to a variety of robustness checks. drops a large proportion of the observations, addressing the concern
Columns 2–4 of Table 4 report the results of the specifications that the sample composition changes over the study period. In all cases,
that control for differential time trends flexibly. First, I replace time the estimated effects are similar to (or slightly larger than) the baseline
fixed effects with village-by-year dummies, absorbing shocks that are estimates.
common to all households and individuals in the same village-year Last, I conduct two sets of placebo tests. Using non-political trust
(column 2). Compared to the baseline estimates, the effects on both variables as placebo outcomes, I provide a set of placebo tests for
outcomes shrink, but remain economically large and statistically signif- political trust. In the CFPS, other variables measuring social trust are
available: general trust and individuals’ trust towards parents, neigh-
icant. Second, I add a village-specific linear time trends, which does not
bors, Americans, strangers and doctors. I regress these non-political
appreciably change the effect magnitude (column 3). Finally, I control
trust variables on exposure to land expropriation. Table 5 presents
for county-by-year fixed effects (column 4). The effects are actually
the results, showing that none of the non-political trust variables are
larger than the ones obtained from column 2.
affected by land expropriation.28 These results lend strong support to
Column 5 of Table 4 further includes two time-varying covariates
the causal link between land expropriation and political trust.
(age and educational-level dummies) in the regressions. The estimated
Following Chen and Yang (2019a), I also conduct one additional
effects are almost the same as the baseline results. The minimal changes
placebo test for both outcomes. Under the null that there is no asso-
suggest that time-varying characteristics are not likely to drive the
ciation between pseudo-exposure and the political outcomes for the
results.
control group, randomly assigned pseudo-exposure (to land expropria-
Column 6 of Table 4 presents the results of the specification, which tion) to control households would not affect the political outcomes. To
includes interaction terms between time dummies and individual-, formally test this logic, I first drop the actual treatment group and re-
household-, and village-level time-invariant characteristics in Table 2. sample a certain proportion of households without replacement in each
The data shows that villages located closer to the county center are year and code them as ‘‘expropriated’’ households (pseudo-treatment
likely to have a earlier expropriation project. While my specification group); the proportions correspond to the percentages depicted in
includes individual fixed effects that account for all time-invariant Fig. 1. The remaining households serve as a new control group. Using
differences due to being closer to the county center, allowing the time the newly constructed sample and pseudo-treatment group, I then
effects to depend on proximity to the county center further controls for reestimate the baseline model at the individual level to obtain the
differences in political outcomes over time that are due to villages with pseudo-treatment effect. I follow this procedure 5,000 times, randomly
earlier expropriation programs being closer to the county center. The assigning pseudo-exposure to land expropriation.
estimated effects are very similar to the ones obtained from the baseline Fig. 4 plots the distribution of t-statistics from the 5,000 esti-
specification (column 1). mated pseudo-treatment effects on political trust (Panel A) and political

27 28
Landesa (2012) suggests that more than 65 percent of the mass conflicts The results are similar if I replace time fixed effects with village-by-year
are related to land in China in 2010. fixed effects (available upon request).

11
W. Sha Journal of Development Economics 160 (2023) 102985

Fig. 4. Distribution of t-statistics. Notes: This figure plots the distribution of t-statistics from the 5,000 estimated pseudo-treatment effects on political outcomes using the baseline
model by randomly assigning pseudo-exposure to land expropriation to control households. The red vertical lines mark the t-statistics from actual exposure to land expropriation.
The reported 𝑝-value is the share of the pseudo-treatment t-statistics that is larger than the actual t-statistics, in absolute value.
Source: China Family Panel Studies.

Fig. 5. Dynamic effects of land expropriation on political outcomes. Notes: Unit of observation is the individual-year. The sample is a four-year panel (2012, 2014, 2016, and
2018) for Panel A and a four-year panel (2010, 2012, 2014, and 2016) for Panel B. The sample excludes individuals who had their land expropriated before 2012 (2010) for
Panel A (Panel B). The shaded region indicates the expropriation was happening during this time period. The CFPS survey is conducted in every two years, thus, the gap between
two consecutive periods is two years (the horizontal axis).
Source: China Family Panel Studies.

conflict (Panel B). I also report the value of the actual t-statistics, I estimate the dynamic effects using Eq. (2) and plot the estimated
corresponding to the treatment effect obtained from the baseline model coefficients and the corresponding 95% confidence intervals in Fig. 5.
using actual exposure to land expropriation. The reported p-values The regression results are reported in Table A.5. Since the main sample
are the share of the pseudo-treatment t-statistics that are larger than is not a balanced panel, I report the results using both unbalanced and
the actual t-statistic in absolute value. One can see from the fig- balanced panels to address the concern that the results could be biased
ure that randomly assigned exposures produce a small (and negligi- by changes in sample composition. There are no effects of being in
ble) proportion of t-statistics in explaining political trust or political the treatment group on political trust in the years before individuals
conflict. are exposed to land expropriation, but right after exposure, it drops
sharply, and then appears to flatten out and eventually attenuate to
5.2. Pre-trends, persistence, and spillovers zero four years later (Panels A and B). A similar pattern is found for
political conflict, with the effects being persistent for a slightly longer
In this subsection, I report results of the event study specification period (Panels C and D). Note that the results are robust to using the
and the tests for spillover effects. These two sets of results not only method proposed by Sun and Abraham (2021).
serve as internal validity checks, but also are important for understand- These results yield two important insights. First, the results provide
ing and interpreting the findings of this paper. evidence that the pretreatment parallel trends assumption, which is a

12
W. Sha Journal of Development Economics 160 (2023) 102985

Table 6
Spillover effects of land expropriation on political outcomes.
Source: China Family Panel Studies.
Control sample
Trust Conflict Trust Conflict
(1) (2) (3) (4)
Land expropriation −0.3485∗ 0.0016
(0.2008) (0.0222)
County expropriation (share) 0.4426 0.0862
(0.6415) (0.0779)
Ratio of treated villagers −0.1996 0.0177
(0.3229) (0.0306)
Land expro./Ratio mean 0.0433 0.0456 0.1427 0.1219
Land expro./Ratio SD 0.1875 0.1928 0.1571 0.1484
Dep. var. mean 5.2788 0.0486 5.3130 0.0462
Dep. var. SD 2.6619 0.2151 2.6577 0.2099
# of villages 451 454 451 453
# of observations 70937 69748 61308 57624
Adj. R-squared 0.3211 0.1442 0.3170 0.1344
Individual fixed effects Yes Yes Yes Yes
Time fixed effects Yes Yes Yes Yes

Notes: Unit of observation is the individual-year. The sample is a four-year panel (2012, 2014, 2016, and 2018) for columns 1 and
3 and a four-year panel (2010, 2012, 2014, and 2016) for columns 2 and 4. Columns 3–4 only use the control sample. Movers and
always-treated individuals are excluded in the regressions for all columns. Trust towards local government officials is a categorical
variable ranged from 0 (extremely low trust) to 10 (extremely high trust) in a given year. Conflict with local government officials is a
dummy that equals to 1 if the individual had conflicts with local government officials in the past year for the 2012, 2014, and 2016
waves, and equals to 1 if the individual has ever had conflicts with officials in history for the 2010 wave. Land expropriation is a
dummy that is 1 if the individual is in a household whose land has ever been expropriated in a given year. Country expropriation is a
percentage that equals to leave-self-out share of individuals who had their land expropriated in the county. Ratio of treated villager is
the share of individuals subject to land expropriation in the village. Robust standard errors in parentheses are clustered at the village
(of current residence) level. ∗∗∗ : significant at 1%; ∗∗ : significant at 5%; ∗ : significant at 10%.

critical underlying assumption for the difference-in-differences frame- 5.3. Effects on additional outcomes
work, holds for both outcomes. Before exposure to land expropriation,
the difference in political trust as well as in political conflict between Up to now, I focus attention on outcomes related to local gov-
expropriated (treatment) and non-expropriated (control) individuals is ernment officials. Can the adverse impacts extend to other political
not statistically significant from zero, which suggests that pre-trends do dimensions? How do those whose land is expropriated assess the gov-
not differ between treatment and control groups. Second, the effects on ernment? In this subsection, I answer these questions by studying how
both outcomes are not persistent over a long period of time. The lack of individuals who are subject to land expropriation assess government
persistence could reflect declining salience of post events, especially if service, performance, and quality. To do so, I examine four additional
local leaders change over time. It could also in part be due to improved outcomes. The first is unfair treatment by local officials measured
living conditions over time in villages with land expropriation due by whether an individual experiences being unfairly treated by local
to better infrastructure conditions and more public goods provisions. government officials in the past year. The second is unreasonable delay
Those subject to land expropriation could gradually become more at local agencies, referring to when citizens come to local government
supportive of the government after witnessing improvements of living agencies to ask for help but local officials will shirk responsibility and
conditions. ask them to go to other agencies for help (known as ‘‘kicking balls’’
In my empirical context, spillovers from the expropriated to the
in China). It is a dummy variable that is 1 if an individual reports
non-expropriated are also worth exploring. On the one hand, if the
such experience in the past year and 0 otherwise. The third outcome
potential outcomes of the control group of individuals whose land was
is individuals’ evaluation of the performance of the county government
not expropriated is affected by the expropriation of the land of their
in the past year compared to the performance in earlier years, which
neighbors, the baseline estimates would be biased towards zero. On
is a categorical variable ranging from 1 to 5 (much worse, worse,
the other hand, the existence of spillovers also would suggest that the
the same, better, and much better). Last, I examine the effect of land
baseline estimates understate the treatment effects, that is to say, the
expropriation on individuals’ perceptions of the severity of nationwide
total political costs are larger with spillovers. Again, as discussed above,
I did not hold strong priors on the existence of or direction of spillovers. corruption, a categorical variable ranging from 1 (extremely low) to 10
Table 6 reports the results of spillover effects. In columns 1–2, I (extremely high).
augment Eq. (1) with the share of individuals whose land was expropri- I re-estimate Eq. (1) by placing these four outcomes on the left-hand
ation (excluding oneself) within counties, while I regress the outcomes side. Table 7 reports the estimated effects. Columns 1–4 show that indi-
on the ratio of treated villagers within villages using the control sample viduals who are subject to land expropriation are more likely to report
that excludes treated units in columns 3–4. In all columns, I control for being unfairly treated by local officials as well as being unreasonably
individual and time fixed effects, the preferred set of fixed effects in the delayed at local agencies, with effects being both statistically significant
regression analyses throughout the paper. Overall, I do not find any and economically sizable. The effect on individuals’ evaluation of the
statistically significant results, suggesting no spillover effects. Recall performance of the county government is not statistically different
that my baseline estimates survive when using specifications including from zero (columns 5–6). The effects on individuals’ perceptions of the
village-by-year or county-by-year fixed effects (see columns 2 and 4 severity of nationwide corruption also is remarkable (columns 7–8). I
of Table 4), in which I compare the individual-level outcomes across also check for parallel pre-trends using the event study specification
the treatment and control groups within the same village (or county) and plot the results in Figs. A.4 and A.5, showing no evidence of
and the same year, suggesting that there is at least no sufficiently the existence of differential pre-trends.29 Taken together, the evidence
large spillovers across individuals within villages or counties. As noted
earlier, this could reflect positive and negative spillovers canceling each
29
other out. See Table A.6 for the corresponding regression results.

13
W. Sha Journal of Development Economics 160 (2023) 102985

Table 7
Effects of land expropriation on additional political outcomes.
Source: China Family Panel Studies.
Unfair treatment Unreasonable delay Evaluation of performance Perception of severity
by local officials at local agencies of county government of nationwide corruption
(1) (2) (3) (4) (5) (6) (7) (8)
Land expropriation 0.0348∗∗∗ 0.0147 0.0361∗∗∗ 0.0207∗∗ −0.0424 −0.0235 0.1852∗ 0.0906
(0.0120) (0.0096) (0.0135) (0.0104) (0.0289) (0.0221) (0.1070) (0.0793)
Land expro. mean 0.0733 0.0733 0.0737 0.0737 0.0902 0.0902 0.0683 0.0683
Land expro. SD 0.2606 0.2606 0.2613 0.2613 0.2865 0.2865 0.2523 0.2523
Dep. var. mean 0.1205 0.1205 0.1488 0.1488 3.4743 3.4743 6.3403 6.3403
Dep. var. SD 0.3255 0.3255 0.3559 0.3559 0.9258 0.9258 2.8883 2.8883
# of villages 1501 1501 1502 1502 2047 2047 2030 2030
# of observations 73445 73445 72920 72920 92598 92598 70693 70693
Adj. R-squared 0.2007 0.2399 0.2179 0.2660 0.2224 0.2831 0.2196 0.2671
Individual fixed effects Yes Yes Yes Yes Yes Yes Yes Yes
Time fixed effects Yes No Yes No Yes No Yes No
Village FEs X time FEs No Yes No Yes No Yes No Yes

Notes: Unit of observation is the individual-year. The sample is a four-year panel (2010, 2012, 2014, and 2016) for columns 1–4, a five-year
panel (2010, 2012, 2014, 2016, and 2018) for columns 5–6, and a four-year panel (2012, 2014, 2016, and 2018) for columns 7–8. Always-treated
individuals are excluded in the regressions for all columns. Unfair treatment refers to individual’s experience of being unfairly treated by local
government officials, which is a dummy variable that is 1 if the individual had such experience in the past year and 0 otherwise. Unreasonable
delay (stalling) refers to that when citizens come to local government agencies, government officials will shirk responsibility rather than helping
them (‘‘kicking balls’’), which is a dummy variable that is 1 if the individual had such experience in the past year and 0 otherwise. Evaluation
of performance refers to individual’s evaluation of county government performance in the past year, which is ranged from 1 to 5 (much worse,
worse, the same, better, much better). Perception of corruption refers to individual’s perception of the severity of nationwide government
corruption in China, which is ranged from 0 (extremely low) to 10 (extremely high). Land expropriation is a dummy that is 1 if the individual
is lived in a household whose land has ever been expropriated in a given year. Robust standard errors in parentheses, clustered at the village
(of current residence) level. ∗∗∗ : significant at 1%; ∗∗ : significant at 5%; ∗ : significant at 10%.

suggests that individuals whose land was expropriated negatively assess threat to promotion of local political leaders in which social order
government service and quality. plays a veto role. The definition of political conflict is not clear in
the CFPS, presumably ranging from having disputes with officials to
5.4. Discussion on the political impacts organizing/participating in large-scale protests. Recall that the baseline
estimate of the impact on political conflict is 2.5%, which means that
In this subsection, I discuss the economic significance of the es- land expropriation increases the likelihood of experiencing conflicts
timated political impacts and the political implications of the lack with local officials by about 50%. As noted by Scott (1985), affected
of persistence and spillovers. For impacts on political trust, I find farmers may use low-profile techniques to express resistance frequently
evidence of heterogeneous treatment effects (see detailed discussion (‘‘everyday forms of peasant resistance’’), rather than participating in
in next section). Take the estimate presented in column 1 of Panel well-organized large-scale protest activities that are dangerous, if not
A of Table 9 as an example, the effect size amounts to around 0.28 suicidal, for peasants.31 This view is in part supported by the results
standard deviations in badly governed villages, which is more than that individuals who have their land expropriated negatively assess
four-time larger than the baseline estimate (equivalent to 0.07 standard government service and quality. Multiplied many thousand-fold, these
individual acts of resistance may reduce the effectiveness of policy
deviations). To benchmark these estimates, I now compare them with
implementation at the local level.
the impact of the Great Leap Forward (GLF), a life-changing historical
My empirical results show that the adverse political effects attenuate
trauma. Chen and Yang (2019a) use the 2010–2012 CFPS data (a
to zero after about four years and do not spill over to the neighbors
subsample of mine) to study the effects of the GLF on political trust
not subject to land expropriation. The lack of persistence suggests that
(the same outcome as mine). Their estimate of the adverse impact
land expropriation has political costs only in the short run. The lack
of having experienced the GLF is about 0.76 standard deviations for
of spillovers implies that local governments may face fewer obstacles
famine survivors in harder hit regions.30 My baseline estimate and
when eliciting compliance with the land expropriation projects from
estimate for villages with bad governance are equivalent to about 9%
local residents, especially for projects that generate public benefits and
and 28% of the estimate of the famine impact, respectively. This simple
thus could also gain support from those affected. Taken together, these
comparison suggests that the impacts of land expropriation on political
results suggest that the adverse political impacts are not persistent and
trust are significant given that having one’s land expropriated is much
restrictive to affected individuals only. Put differently, to some extent,
less of a life-changing event than experiencing the GLF, one of the worst land expropriation in rural areas may not pose a fatal threat to the
famines in human history that led to excess deaths of at least 30 million Chinese political regime.
people.
As for the impacts on political conflict, I do not find strong evi- 6. Political and economic conditions
dence of heterogeneous treatment effects. The sample average of the
occurrence of political conflict is 5%, indicating that one in twenty This section investigates the circumstances in which political costs
individuals aged 15 or above in rural China has had conflicts with emerge as well as the conditions under which the adverse political
local officials. This high frequency makes it unlikely that most of effects may be mitigated, focusing in particular on compensation fair-
the described conflicts are peasant protests, which could pose a fatal ness, governance quality, project benefits, and agricultural dependence.

30
See column 1 of Panel A of Table 1 in Chen and Yang (2019a): 1.476+0.434
2.506
≈ 31
Note that up to now I have not considered the reallocation programs,
0.76, the largest estimate among all baseline estimates they provided in the such as housing demolition, which could make affected individuals homeless
table. Although they use a very restrictive sample, standard deviations of and are more likely to trigger large-scale protests if they are not compensated
political trust in both my paper and theirs are around 2.5. fairly.

14
W. Sha Journal of Development Economics 160 (2023) 102985

These conditions need not be mutually exclusive. Indeed, they might that land expropriation does create a persistent negative shocks to
be interrelated. It is worth noting first that compensation is not so household agricultural income and land assets. This is not surprising as
low that land expropriation creates large, negative economic shocks to both variables are directly associated with the amount of land allocated
affected households. But it is very low when compared to the market to the household. In addition, households who have had their land
value of the land, which in turn may lead to political costs. Additionally expropriated are more likely to have at least one household member
and relatedly, how local governments manage and justify the projects, who migrated (Panel C),35 and earn more wage income (Panel D).
whether local households perceive public benefits from the programs, However, I find that there is almost no effect of land expropriation on
and to what extent the households rely on the agricultural land also household total income or food consumption (Panels E and F). Finally,
may be important in determining the political effects.
I do not find strong evidence that household savings increase signif-
icantly (Panel G). Although land expropriation does enhance saving
6.1. Compensation fairness
assets right after treatment, the effects are not persistent.
The compensation received by households for land expropriation Overall, land expropriation does create negative shocks to agricul-
could play an important role in determining the political impacts. tural income and land assets persistently, and the lack of significant
However, the CFPS does not collect detailed compensation informa- increase in savings suggests that the lost land wealth is not being
tion.32 As I argued above, the difference between the market price fully offset by compensation. However, total household income re-
and the price of expropriated land is so large that variation in actual mains unchanged, likely due to the ability of households to cope with
compensation across households may not adequately reflect percep- these shocks by migrating and earning more wage income. Migration
tions of fairness of the compensation if households compare their of course could involve substantial disutility of living in cities and
compensation with the market value of the land. Additionally, the risks that are not captured by the total income and saving measures.
actual compensation that households receive could depend on their Taken together, these results suggest that there are some lost economic
interactions with local government officials. Thus, having conflicts with well-being for households whose land was expropriated.
local officials (e.g., collective action taken by villagers) may increase One may ask how we should reconcile the seemingly conflicting
the compensation amount if local officials do not want to slow down results that land expropriation does not have large negative economic
the expropriation process, or they are concerned about political risks impacts but does have adverse political consequences. As discussed
caused by social unrest triggered by land expropriation.
above, I argue that it is the market value of the land that serves as a
I employ two indirect ways to study the role of compensation in
benchmark for households to assess the fairness of compensation. When
determining the political impacts of land expropriation. First, I examine
the expropriated price of the land paid to households is much lower
the economic impacts of land expropriation at the household level.
than the land price charged by the government in the primary market,
Generally speaking, most rural households rely on their agricultural
it suggests that local governments extract rents from local farmers.
land to generate income and produce food. When land is seized and
the compensation is low, it may create a negative economic shock to Given that information on the market price of land is publicly available,
affected households, which may influence the political reaction. Hence, households whose land was expropriated are aware of the market value
it is important to investigate the magnitude of these economic shocks. of land.
To this end, I investigate how land expropriation changes households’ Next, I compare land expropriation with housing demolition to
economic behavior and outcomes. I first examine the impacts on two study the perceived unfairness of compensation. These two policies
outcomes that I expect to be directly affected by land expropriation: are comparable in the sense that both are common types of eminent
agricultural income and land assets. I then examine how land expro- domain in China, in which the government takes properties (land and
priation affects household members’ propensity to migrate (at least one housing, respectively) away from local residents. Importantly, in the
household member is migrated),33 and whether it increases household case of housing demolition, in addition to monetary compensation,
wage income. Next, I investigate the impacts on total household income most affected households are compensated with new houses (apart-
and food consumption. Last, I study the effects on household savings. ments) and/or another piece of homestead. Chen (2014) provides a
Importantly, household savings are expected to increase (and perhaps legal analysis of China’s bifurcated land system with urban land rights
persistently) if the compensation is large enough, given that the Chinese being better protected. This bifurcation was enlarged in 2011 (the
households generally have high saving rates and about 71% of house-
beginning of the study period of this paper) when the Chinese central
holds who had their land expropriated receive cash as compensation.
government passed new regulations that require local governments to
I re-estimate Eq. (2) by replacing the political outcomes with
take into account of market value of the property based on assessments
household-level economic outcomes, and I replace individual fixed
of an independent third-party when determining the compensation
effects with household fixed effects. In the regressions, the migration
for affected households.36 Unfortunately, these regulations are only
variable is a dummy equal to 1 if the household has at least one member
who migrated in the past and 0 otherwise. All of the other household applicable to expropriation of urban land (i.e., housing demolition),
outcomes are continuous variables, all of which are normalized by but not to expropriation of rural agricultural land (see Chen (2014)
family size and logged. for more detail). Given that the compensation package is much more
I plot the estimated effects on household-level outcomes using generous, housing demolition often is viewed as a large positive wealth
the event study specification in Fig. 6 and report the corresponding shock. For example, using the same data set, Sha and Zou (2022)
regression results in Table A.7.34 From Panels A and B, one can see find that households whose house was demolished tend to have more
housing value and wealth, consume more, and work less. In such
a context, there should be no adverse (or even positive) effects on
32
The CFPS provides total compensation received by households, however, political outcomes.
it does not provide expropriated acreage, so I cannot calculate the expropriated
price of agricultural land. Using total compensation alone will be misleading.
33
This variable is constructed based on whether one household receives
35
remittance in a given year or not. The CFPS does provide information about Using a two-year (2013 and 2015) panel constructed from the CHFS, Ma
whether one work in the non-agricultural sector, but does not contain detailed and Mu (2020) also found that land expropriation increases individual
information on migration. migration rate by 4.5–6.8 percentage points.
34 36
The dynamic estimates obtained using the method proposed by Sun and See the Regulations for Expropriation and Compensation for Houses on
Abraham (2021), which remain largely robust, are plotted in Fig. A.6. State-owned Land of 2011 for morel detail.

15
W. Sha Journal of Development Economics 160 (2023) 102985

Fig. 6. Dynamic effects of land expropriation on household outcomes. Notes: Unit of observation is the household-year. The sample is a five-year household panel (2010, 2012,
2014, 2016, and 2018) and excludes households who had their land expropriated before 2010. The shaded region indicates the expropriation was happening during this time
period. The CFPS survey is conducted in every two years, thus, the gap between two consecutive periods is two years (the horizontal axis).
Source: China Family Panel Studies.

To test this empirically, I replace the land expropriation dummy does not bring a large negative income shock to affected households,
in Eq. (1) with an indicator of whether the individual is in a household the perceived unfairness of below-market compensation may greatly
whose housing has been demolished in a given year. Table 8 presents contributes to generating adverse political impacts.
the estimated impacts: columns 1–2 report the results using the main
sample used in this paper (i.e., the rural sample); columns 3–4 report 6.2. Governance quality
the results only using the urban sample, which is most relevant (be-
cause urban areas experienced more housing demolition in China) and It is crucial how land expropriation is administered by local gov-
has a less heterogeneous control group; columns 5–6 report the results ernments, which largely reflects the quality of governance at the local
using the sample including both rural and urban areas. Consistent with level. In China, local government officials generally are not accountable
the above conjuncture, most of the effects of housing demolition on downward to citizens but rather respond to promotion incentives deter-
political outcomes are statistically significantly different from zero; and mined by upper-level government leaders. Consequently, the welfare
the estimated coefficient is marginally statistically significant (column of households often is of second-order importance for local political
6 of Panel A) when comparing political trust of individuals with and leaders who tend to be radical and aggressive in implementing new
without land expropriation within the same village and the same year. projects. One common practice of local governments is to make the
Taken together, the evidence suggests that although land expropriation expropriation procedure less transparent and deploy local police to

16
W. Sha Journal of Development Economics 160 (2023) 102985

Table 8
Effects of housing demolition on political outcomes.
Source: China Family Panel Studies.
Trust towards local government officials
Only urban Include urban
(1) (2) (3) (4) (5) (6)
Panel A: Effects on political trust
Housing demolition 0.0613 0.1540 0.0035 0.3134 0.0800 0.1868∗
(0.1666) (0.1230) (0.2005) (0.2109) (0.1300) (0.1059)
Housing demo. mean 0.0216 0.0216 0.1666 0.0381 0.0255 0.0255
Housing demo. SD 0.1454 0.1454 0.3726 0.1913 0.1576 0.1576
Dep. var. mean 5.2204 5.2204 4.6621 4.6745 5.0808 5.0808
Dep. var. SD 2.6643 2.6643 2.4904 2.4910 2.6344 2.6344
# of villages 2080 2080 1252 1193 3630 3630
# of observations 89743 89743 32760 28383 119782 119782
Adj. R-squared 0.3254 0.3413 0.3700 0.3910 0.3391 0.3558
Conflict with local government officials
Only urban Include urban
(1) (2) (3) (4) (5) (6)
Panel B: Effects on political conflict
Housing demolition 0.0016 0.0035 0.0188 0.0097 0.0109 0.0075
(0.0173) (0.0124) (0.0186) (0.0203) (0.0129) (0.0105)
Housing demo. mean 0.0221 0.0221 0.0281 0.0281 0.0235 0.0235
Housing demo. SD 0.1469 0.1469 0.1652 0.1652 0.1516 0.1516
Dep. var. mean 0.0495 0.0495 0.0446 0.0446 0.0482 0.0482
Dep. var. SD 0.2169 0.2169 0.2064 0.2064 0.2141 0.2141
# of villages 1520 1520 934 934 2702 2702
# of observations 81523 81523 28044 28044 110478 110478
Adj. R-squared 0.1377 0.1732 0.1833 0.2126 0.1480 0.1807
Individual fixed effects Yes Yes Yes Yes Yes Yes
Time fixed effects Yes No Yes No Yes No
Village FEs X time FEs No Yes No Yes No Yes

Notes: Unit of observation is the individual-year. The sample is a four-year panel (2012, 2014, 2016, and 2018) for Panel A, and a
four-year panel (2010, 2012, 2014, and 2016) for Panel B. Columns 1–2 use the main sample; columns 3–4 use urban villages only;
columns 5–6 add urban villages to the main sample. Always-treated individuals are excluded in the regressions for all columns. Trust
towards local government officials is a categorical variable ranged from 0 (extremely low trust) to 10 (extremely high trust) in a
given year. Conflict with local government officials is a dummy that equals to 1 if the individual had conflicts with local government
officials in the past year for the 2012, 2014, and 2016 waves, and equals to 1 if the individual has ever had conflicts with officials
in history for the 2010 wave. Land expropriation (housing demolition) is a dummy that is 1 if the individual is in a household whose
land (housing) has ever been expropriated (demolished) in a given year. Robust standard errors in parentheses are clustered at the
village (of current residence) level. ∗∗∗ : significant at 1%; ∗∗ : significant at 5%; ∗ : significant at 10%.

force citizens to comply with the expropriation demands.37 Relatedly, the 2014 wave of the CFPS.38 Slightly more than 60% of villages report
access to information can increase the quality of public goods and that they post information about land expropriation and other major
service (e.g., Lewis-Faupel et al., 2016), and local monitoring also can events on bulletin boards (Fig. 7, Panel A). To be more precise, I only
hold public workers accountable (e.g., Björkman and Svensson, 2009). use observations in 2014 or afterwards and so drop all individuals
Additionally, corruption and other forms of misconduct could reduce who have their land expropriated before 2014. That is, I use the 2014
the quality of public goods delivered by the government (e.g., Reinikka wave as baseline and examine whether the political impacts differ
and Svensson, 2004; Olken, 2007). Indeed, corruption is not uncommon with respect to this baseline characteristic, restricting attention only
in the land market in China (Chen and Kung, 2019). The Wukan event to individuals who have been expropriated in 2014 or later.
vividly exemplifies how local officials are able to use public office for Second, I measure the degree of government transparency at the city
private gains. level using an index from Nie et al. (2019). This transparency index is
To study the role of the quality of local governance in mitigating the a standardized score based on the degree of administrative information
adverse political impacts of land expropriation, I interact the treatment disclosure and the degree of fiscal information disclosure. Based on
indicator with variables that capture the quality of governance across the mean level of the index, I create a dummy indicating high (above-
regions (villages and cities). The first variable is whether the village mean) government transparency. Next, I turn to whether the adverse
committee posts information about the land expropriation projects on political impacts of land expropriation can be mitigated by less corrupt
the bulletin board, which makes governance more transparent and governments. My third variable is a corruption index, also provided
avoids the abuse of power. Data on this variable is only available in

38
Up to now, the CFPS conducted village-level surveys only in 2010 and
37
See, for example, Lucy Hornby, ‘‘China Migration: Dying for Land’’, Fi- 2014. In the 2010 village-level survey, the CFPS did not ask about whether
nancial Times, August 7, 2015. Available online: https://www.ft.com/content/ information regarding land expropriation is available for villagers from the
33ae0866-3098-11e5-91ac-a5e17d9b4cff. Last access on August 17, 2020. bulletin board.

17
W. Sha Journal of Development Economics 160 (2023) 102985

Fig. 7. Village committee bulletin contents and village infrastructure improvement. Notes: Panel A plots the share of villages that report the village committee posts the following
contents on the bulletin board when surveyed in 2014: financial conditions, staff recruitment, nomination of village leaders, government policy, disposal of collective assets, major
village events (land expropriation, housing demolition, and re-construction etc.), and implementation of family planning policy. Panel B plots the share of villages that have
had access to the following infrastructure before 2010: electricity, cable radio, cable/satellite TV, postal service, telephone, cellphone signal/service, roads, railway, tap water,
and pipeline gas. Panel C plots the share of villages that have had access to the following infrastructure from January 1st, 2010 to December 31st, 2013: electricity, cable radio,
cable/satellite TV, postal service, telephone, cellphone signal/service, roads, railway, tap water, pipeline gas, public bus, and subway.
Source: China Family Panel Studies.

by Nie et al. (2019), which is a standardized score based on the fee by the baseline model are most likely to be driven by regions with
for obtaining a food safety permit from a local government bureau worse governance quality. In line with the literature, these findings
and the total number of corruption-related news on Baidu (the largest highlight the importance of sound local governance in implementing
Chinese search engine).39 I also create a dummy indicating low (below- development projects in China (e.g., Park and Wang, 2010; He and
mean) government corruption. These two indices rank the degree of Wang, 2017; Wong et al., 2017; Bardhan, 2020).
government transparency and corruption in Chinese cities (Beijing,
Tianjin, Shanghai, Chongqing, and all prefectures) in 2018, which is 6.3. Project benefits
the last year of my data. I will keep all observations for regression
analysis and assume that the cross-sectional variation in each of these In theory, local residents could favor land expropriation projects if
two indicators persists over time. they support projects that create benefits for them.42 In this case, the
The main concern in this exercise is that individuals in regions interests of local citizens and the government are aligned. As I have
with worse quality of governance may also experience shocks that are shown in Section 2, most of the expropriated agricultural land is used
correlated with the timing of land expropriation and the time path for public projects such as highway and railway construction, but at the
of political outcomes, but this does not occur in regions with better same time, there is a nontrivial proportion of other (private) projects
governance. To control flexibly for community-specific shocks, I will that are not necessarily beneficial for local households. The interests of
allow the time fixed effects to be specific to each village. local citizens and the government are more likely to be aligned when
Table 9 reports the estimated impacts obtained from regressions local governments carry out public programs that benefit local residents
including the interaction terms.40 In theory, one should expect that the and improve their quality of life. In this circumstance, the political costs
coefficient on the interaction term is statistically significantly different may be minimized. Relatedly, individuals who benefit from government
from zero and its sign is opposite to the sign of the coefficient on programs are more likely to support the government (e.g., Manacorda
the treatment dummy. Consistent with this conjuncture, I find that the
et al., 2011; Huet-Vaughn, 2019). Thus, one should expect that the
adverse impacts on political trust can be mitigated when villages make
adverse political effects of land expropriation are more pronounced if
information about expropriation publicly available (Panel A, columns
the expropriated land is not used for public projects.
1 and 3), and when cities exhibit high government transparency (Panel
The CFPS data does not contain information on the usage of ex-
B, columns 1 and 3) and low levels of government corruption (Panel C,
propriated agricultural land, but it does provide information about
columns 1 and 3). As for the effects on political conflict, I do not find
whether or not the village has had access to particular types of infras-
evidence for a mitigating role of better local governance. Nevertheless,
tructure before 2010 or during 2010–2013. Fig. 7 depicts the share
one can observe that the impacts on political conflict are only statis-
of villages that had access to various kinds of infrastructure before
tically significant and much larger in regions with better governance
2010 (Panel B) and during 2010–2013 (Panel C). One can see that
when I conduct a subgroup analysis and control for individual and time
although most villages already had access to basic infrastructure before
fixed effects (see Panel A of Table A.8).41 Taken together, these results
2010, there is a considerable proportion of villages that began to have
suggest that the adverse political impacts of land expropriation found
access to basic infrastructure during 2010–2013 (more precisely, from 1
January 2010 to 31 December 2013). For example, among the sampled
39 villages, about 18% (21%) of villages began to have access to roads (tap
The former is indicative of how local governments issue business permits
to firms: only charge a reasonable service fee or charge a much higher fee that water) during this period.
suggests the existence of rent-seeking.
40
I report the estimated impacts using different subsamples in Table A.8.
41 42
Table A.9 reports the results of a regression interacting the treatment Related to governance quality, at the local level, poorly governed villages
dummy with both high government transparency and low levels of government may focus less on projects that provide public goods and may have more poorly
corruption, the findings are similar. designed projects which are less likely to be successful (e.g., Khwaja, 2009).

18
W. Sha Journal of Development Economics 160 (2023) 102985

Table 9
Effects of land expropriation on political outcomes: Governance quality.
Source: China Family Panel Studies.
Trust Conflict Trust Conflict
(1) (2) (3) (4)
Panel A: Bulletin information about expropriation in 2014
Land expropriation −0.7291∗∗∗ 0.0702 −0.6233∗∗∗ 0.0518∗
(0.2267) (0.0539) (0.1657) (0.0300)
Land expro. X bulletin info. available 0.5794∗∗ −0.0327 0.5270∗∗ −0.0396
(0.2835) (0.0697) (0.2041) (0.0358)
Land expro. + land expro. X bulletin info. available −0.1497 0.0374 −0.0962 0.0123
(0.1723) (0.0442) (0.1192) (0.0196)
# of villages 402 401 402 401
# of observations 50278 32597 50278 32597
Adj. R-squared 0.3368 0.1879 0.3541 0.2022
Panel B: Government transparency in 2018
Land expropriation −0.3680∗∗∗ 0.0313∗ −0.3041∗∗∗ 0.0162
(0.1292) (0.0165) (0.0902) (0.0122)
Land expro. X high transparency 0.4023∗∗ −0.0126 0.3162∗∗ −0.0124
(0.1779) (0.0203) (0.1446) (0.0159)
Land expro. + land expro. X high transparency 0.0344 0.0187 0.0120 0.0039
(0.1249) (0.0120) (0.1130) (0.0103)
# of villages 1150 917 1150 917
# of observations 67007 64313 67007 64313
Adj. R-squared 0.3264 0.1507 0.3418 0.1850
Panel C: Government corruption in 2018
Land expropriation −0.3038∗∗ 0.0354∗∗ −0.2833∗∗∗ 0.0074
(0.1367) (0.0141) (0.1069) (0.0109)
Land expro. X low corruption 0.2596 −0.0207 0.2562∗ 0.0040
(0.1833) (0.0197) (0.1449) (0.0157)
Land expro. + land expro. X low corruption −0.0442 0.0146 −0.0271 0.0114
(0.1251) (0.0140) (0.0978) (0.0114)
# of villages 1150 917 1150 917
# of observations 67007 64313 67007 64313
Adj. R-squared 0.3263 0.1508 0.3418 0.1850
Individual fixed effects Yes Yes Yes Yes
Time fixed effects Yes Yes No No
Village FEs X Time FEs No No Yes Yes

Notes: Unit of observation is the individual-year. In Panel A, the sample is a three-year panel (2014, 2016, and 2018) for
columns 1 and 3, and a two-year panel (2014 and 2016) for columns 2 and 4. In Panels B and C, the sample is a four-year
panel (2012, 2014, 2016, and 2018) for columns 1 and 3, and a four-year panel (2010, 2012, 2014, and 2016) for columns 2
and 4. Always-treated individuals are excluded in the regressions for Panels B and C. Trust towards local government officials
is a categorical variable ranged from 0 (extremely low trust) to 10 (extremely high trust). Conflict with local government
officials is a dummy that equals to 1 if the individual had conflicts with local government officials in the past year for the
2012, 2014, and 2016 waves, and equals to 1 if the individual has ever had conflicts with officials in history for the 2010
wave. Land expropriation is a dummy that is 1 if the individual is in a household whose land has ever been expropriated in
a given year. ‘‘bulletin info. available’’ equals to 1 if the individual is in a village that reports the village committee posts
information about land expropriation projects on the bulletin board in 2014, and 0 otherwise; ‘‘high transparency’’ equals
to 1 if the individual is in a city whose transparency level is above the mean in 2018, and 0 otherwise; ‘‘low corruption’’
equals to 1 if the individual is in a city whose corruption level is below the mean in 2018, and 0 otherwise. Robust standard
errors in parentheses are clustered at the village (of current residence) level. ∗∗∗ : significant at 1%; ∗∗ : significant at 5%; ∗ :
significant at 10%.

I label the latter set of villages (i.e., the ones in Panel C of Fig. 7) as expropriation in the same period, and then examine whether or not
villages with infrastructure improvement during 2010–2013, and make individuals in villages with or without infrastructure improvement
use of them by assuming that land expropriation projects implemented during 2010–2013 respond to land expropriation differently. Notice
in these villages during 2010–2013 generated public benefits.43 Due to that villages without infrastructure improvement during 2010–2013
limited data, I restrict the sample to a two-year panel (2012 and 2014). include those that already had access to basic infrastructure before
I also drop individuals whose land was expropriated before 2010.44 By 2010.
doing so, I focus attention on villages that experienced infrastructure Table 10 presents the estimated impacts using an interaction term
improvement during 2010–2013 and individuals who experienced land as for earlier tests.45 Again, I expect the coefficient on the interaction
term to be statistically significantly different from zero and the sign
to be the opposite of the coefficient on the uninteracted treatment
43
Villages with infrastructure improvement during 2010–2013 are the ones variable. Consistent with this hypothesis, I find that the effects on
that have had access to at least one of the following infrastructure from Jan- political trust is mitigated in villages with infrastructure improvement
uary 1st, 2010 to December 31st, 2013: electricity, cable radio, cable/satellite during 2010–2013, whether or not I control for village-year fixed time
TV, postal service, telephone, cellphone signal/service, roads, railway, tap
effects (columns 1 and 3). Although this is not significant for the effects
water, pipeline gas, public bus, and subway.
44 on political conflict, it is worth noting that for political conflict the
Recall that in the CFPS, the information on political trust is not available
in the 2010 wave. While the information on political conflict is available in signs of the coefficients also are opposite as expected. Taken together,
the 2010 wave, it is a dummy that is 1 if the individual has ever had conflicts
with local officials in history, which might be related to expropriation events
45
that happened before 2010. Thus, I drop the 2010 wave in the regressions in The estimated impacts using different subsamples are reported in
which political conflict is the outcome. Table A.10.

19
W. Sha Journal of Development Economics 160 (2023) 102985

Table 10
Effects of land expropriation on political outcomes: Project benefits.
Source: China Family Panel Studies.
Trust Conflict Trust Conflict
(1) (2) (3) (4)
Land expropriation −0.6262∗∗ 0.0058 −0.4313∗∗∗ 0.0061
(0.2828) (0.0220) (0.1511) (0.0155)
Land expro. X infrastructure improvement 0.7364∗ −0.0067 0.4399∗ −0.0110
(0.3984) (0.0412) (0.2347) (0.0251)
Land expro. + land expro. X infrastructure improvement 0.1102 −0.0010 0.0086 −0.0049
(0.2823) (0.0349) (0.1797) (0.0198)
# of villages 402 402 402 402
# of observations 38398 36372 38398 36372
Adj. R-squared 0.3444 0.1562 0.3661 0.1731
Individual fixed effects Yes Yes Yes Yes
Time fixed effects Yes Yes No No
Village FEs X Time FEs No No Yes Yes

Notes: Unit of observation is the individual-year. The sample is a two-year panel (2012 and 2014) for columns 1–4. Trust
towards local officials is a categorical variable ranged from 0 (extremely low trust) to 10 (extremely high trust) in a given
year. Conflict with local officials is a dummy that equals to 1 if the individual had conflicts with local government officials
in the past year. Land expropriation is a dummy that is 1 if the individual is in a household whose land has ever been
expropriated in a given year. ‘‘infrastructure improvement’’ is a dummy variable, indicating that the individual is in a village
that has had access to at least one of the following infrastructure from January 1st, 2010 to December 31st, 2013: electricity,
cable radio, cable/satellite TV, postal service, telephone, cellphone signal/service, road, railway, tap water, pipeline gas, public
bus, and subway. Robust standard errors in parentheses are clustered at the village (of current residence) level. ∗∗∗ : significant
at 1%; ∗∗ : significant at 5%; ∗ : significant at 10%.

Table 11
Effects of land expropriation on political outcomes: Agricultural dependence.
Source: China Family Panel Studies.
Trust Conflict Trust Conflict
(1) (2) (3) (4)
Land expropriation −0.3429∗∗∗ 0.0389∗∗ −0.2664∗∗ 0.0334∗∗∗
(0.1252) (0.0157) (0.1044) (0.0118)
Land expro. X non-agricultural employment 0.1865 −0.0212 0.1238 −0.0316∗∗
(0.1886) (0.0179) (0.1684) (0.0152)
Land expro. + land expro. X non-ag. employment −0.1564 0.0176 −0.1427 0.0019
(0.1552) (0.0141) (0.1411) (0.0125)
# of villages 723 630 723 630
# of observations 46188 44732 46188 44732
Adj. R-squared 0.3270 0.1513 0.3480 0.1903
Individual fixed effects Yes Yes Yes Yes
Time fixed effects Yes Yes No No
Village FEs X Time FEs No No Yes Yes

Notes: Unit of observation is the individual-year. The sample is a four-year panel (2012, 2014, 2016, and 2018) for columns
1 and 3, and a four-year panel (2010, 2012, 2014, and 2016) for columns 2 and 4. Always-treated individuals are excluded in
the regressions for all columns. Trust towards local officials is a categorical variable ranged from 0 (extremely low trust) to 10
(extremely high trust) in a given year. Conflict with local officials is a dummy that equals to 1 if the individual had conflicts
with local government officials in the past year. Land expropriation is a dummy that is 1 if the individual is in a household
whose land has ever been expropriated in a given year. ‘‘non-agricultural employment’’ is a dummy variable, indicating that
the individual is in a household that has at least one member working in the non-agricultural sector in the baseline year
(2010). Robust standard errors in parentheses are clustered at the village (of current residence) level. ∗∗∗ : significant at 1%;
∗∗
: significant at 5%; ∗ : significant at 10%.

the results suggest that land expropriation projects which generate becomes less important for the Chinese farmers to earn income. Thus,
development benefits for local residents can reduce or even offset the it is expected that the adverse political effects of land expropriation
political costs of land expropriation. are more salient for the households with all members working in
agriculture prior to having their land expropriated.
6.4. Agricultural dependence To examine the importance of the non-agricultural employment
opportunities in determining the political effects of land expropriation,
I interact the treatment indicator with a variable, indicating whether a
Since agricultural land in rural China is used mainly for generating
household engages in the non-agricultural sector (either self-employed
food and income, to what extent the households depend on agriculture
or hired by others). I use information in the baseline year (i.e., 2010) of
may also be important in determining the adverse political impacts.
the CFPS to construct the variable, which equals one if the household
For example, if the households do not rely heavily on the agricultural has at least one member working in the non-agricultural sector in 2010
sector before experiencing land expropriation, then having their land and zero otherwise. Since I focus only on the information in 2010,
expropriated may not necessarily have adverse political effects. In households newly surveyed in the latter waves are dropped in the
contrast, a household with all adults working in agriculture may be regressions. In the 2010 data, slightly more than 60% of rural house-
sensitive to being exposed to land expropriation. In reality, China has holds have all adult members working in agriculture; for the remaining
been experiencing rapid structural transformation featuring a popula- households, at least one member engages in the non-agricultural sector
tion flow from the agricultural to non-agricultural sectors and from (about 10% have all members working in the non-agricultural sector).
rural to urban areas (e.g., Cai et al., 2008). Given the numerous non- Table 11 presents the estimated effects obtained from regressions
agricultural employment opportunities during this process, agriculture including the interaction term. Here, the coefficient on the interaction

20
W. Sha Journal of Development Economics 160 (2023) 102985

term is expected to be statistically significantly different from zero

and its sign is expected to be the opposite of the coefficient on the Table A.1
Usage of expropriated agricultural land.
uninteracted treatment variable. Consistent with this conjuncture, I find Source: Ma and Mu (2020) (see their Table 1 and Footnote 28), who calculated the
that the adverse political impacts on political conflict can be reduced numbers from the 2013 and 2015 waves of the China Household Finance Survey (CHFS).
when households have members working in the non-agricultural sector During 2013–2015 Before 2013
(column 4). It is noteworthy that the signs of the coefficients in columns (1) (2)
1–3 also are opposite as expected. Furthermore, even though the effects Highway and railway construction 48.15% 33%
on political trust are not statistically significant, we can see that the Housing development 25.93% 33%
impacts on political trust are only statistically significantly different Community infrastructure building 24.07% 8%
from zero and the magnitudes are much larger for households with all Construction land used by enterprises 20.37% –
Construction land used by governments 3.70% –
members working in agriculture in 2010 (see Table A.11).46 Taken to-
Others 9.26% –
gether, the results suggest that the political costs of land expropriation
Notes: The percentages do not add up to one as one household can report multiple
could be reduced when the households depend less (or even little) on
usages. Because the CHFS data contain information on the usage of expropriated land
the agricultural sector before having their land expropriated. is not publicly available to outside users, I directly cite the numbers from Ma and
Mu (2020). Columns 1 and 2 are corresponding to their Table 1 and Footnote 28,
7. Conclusion respectively.

China has been employing land expropriation as a way to build

infrastructure to develop the economy for several decades. On the one
hand, it is clear that land expropriation is not always beneficial for
affected households. Thus, it may occur at the expense of lost political Table A.2
Effects of land expropriation on life satisfaction and depression.
support for the government. On the other hand, the government can
Source: China Family Panel Studies.
provide public goods using the expropriated land or by spending the fis-
Life satisfaction Depression
cal revenue generated from selling expropriated land on public goods. (1) (2)
Therefore, whether land expropriation leads to political costs, and how
Land expropriation −0.0172 0.0264
detrimental are these costs, may depend on the specific context. (0.0232) (0.0251)
This study provides new insight on these questions in the context
Land expro. mean 0.0890 0.0890
of China. By analyzing large-scale individual-level panel data from Land expro. SD 0.2848 0.2847
a nationally representative longitudinal survey in China, I find that # of villages 2076 2129
land expropriation incurs political costs for the Chinese government # of observations 100541 107408
Adj. R-squared 0.2389 0.2750
due mainly to perceived unfairness of compensation. However, the
adverse effects do not persist over many years and do not spill over Individual fixed effects Yes Yes
Time fixed effects Yes Yes
to households whose land is not expropriated. Moreover, the adverse
impacts appear to be driven by regions with worse governance quality Note: Unit of observation is the individual-year. The sample is a five-
year panel (2010, 2012, 2014, 2016, and 2018). Life satisfaction is a
and projects without public benefits. The results shed light on the state’s
categorical variable. Depression is an index constructed from various
trade-off between protecting property rights and pursuing development questions asked in the CFPS survey. Both outcomes are standardized
goals, and also suggest that citizens may be willing to sacrifice prop- within each survey wave (thus, both have a mean of zero and a SD of
erty rights security for potential development benefits. The empirical one). Land expropriation is a dummy that is 1 if the individual is in
findings are not only of interest for better understanding China’s devel- a household whose land has ever been expropriated in a given year.
Robust standard errors in parentheses are clustered at the village (of
opment and politics, but they also have wider policy implications for
current residence) level. ∗∗∗ : significant at 1%; ∗∗ : significant at 5%; ∗ :
developing countries: governments can generate development benefits significant at 10%.
and minimize the political costs by implementing better-designed land
expropriation projects with sound governance that deliver benefits to
local citizens.

Declaration of competing interest

Table A.3
The authors declare that they have no known competing finan- Effect of land expropriation on attrition.
cial interests or personal relationships that could have appeared to Source: China Family Panel Studies.

influence the work reported in this paper. Attrition

(1)

Data availability Land expropriation −0.0025

(0.0112)
# of villages 1013
Data will be made available on request.
# of observations 28455
Adj. R-squared 0.6936
Appendix Wave fixed effects Yes
Village fixed effects Yes
See Figs. A.1–A.6 and Tables A.1–A.11.
Note: Unit of observation is the individual. The panel data is
reduced to cross-sectional data. Attrition is a dummy indicating
whether the individual attrits from the sample or not during the
sample period (2010–2018). Land expropriation is a dummy
that is 1 if the individual is in a household whose land has
ever been expropriated in a given year. Robust standard errors
in parentheses are clustered at the village (of current residence)
level. ∗∗∗ : significant at 1%; ∗∗ : significant at 5%; ∗ : significant
46
The estimated impacts using different subsamples are reported in at 10%.
Table A.11.

21
W. Sha Journal of Development Economics 160 (2023) 102985

Table A.4
The time-invariant determinants of land expropriation.
Source: China Family Panel Studies.
Land expropriation
Individual Household Village
(1) (2) (3) (4) (5) (6)
Intercept 0.3402∗∗∗ 0.3836∗∗∗ 1.0446∗∗∗
(0.0315) (0.0416) (0.0341)
Panel A: Individual-level Characteristics
Male −0.0023 −0.0022
(0.0025) (0.0023)
Han −0.0761∗∗ −0.0079
(0.0327) (0.0179)
CCP membership −0.0234∗∗ 0.0009
(0.0105) (0.0073)
Panel B: Household-level Characteristics
Family genealogy −0.0045 −0.0069
(0.0173) (0.0131)
Distance to nearest high school (km) −0.0422∗∗∗ 0.0093
(0.0107) (0.0098)
Distance to nearest medical clinic (km) −0.0144 −0.0077
(0.0118) (0.0107)
Distance to nearest marketplace (min) 0.0004 −0.0127∗
(0.0098) (0.0075)
Panel C: Village-level Characteristics
Ancestral hall 0.0266 −0.0660
(0.0354) (0.0660)
Any clan with population share >=10% −0.0435 0.0011
(0.0427) (0.0523)
# of clans with population share >=10% 0.0138 0.0055
(0.0091) (0.0088)
Minority area −0.0144 −0.0011
(0.0184) (0.0168)
Natural resource area −0.0008 −0.0103
(0.0126) (0.0115)
Distance to town center (km) −0.0084 0.0072
(0.0189) (0.0217)
Distance to county center (km) −0.0327∗∗∗ −0.0295∗∗
(0.0112) (0.0138)
# of villages/counties 454 454 418 418 141 141
# of observations 33204 33204 9206 9206 457 457
Adj. R-squared 0.0030 0.2485 0.0103 0.2254 0.0102 0.1288
Village fixed effects No Yes No Yes No No
County fixed effects No No No No No Yes

Notes: Unit of observation is the individual for columns 1–2, the household for columns 3–4, and the village for columns 5–6. Land expropriation
is a dummy that is 1 if the individual/household/village has experienced land expropriation in the sample period. At the individual level,
I check for gender dummy, han indicator, and Chinese Communist Party (CCP) membership dummy; at the household level, I check for
family genealogy dummy, the distance between the household’s geographical location to the nearest high school (kilometers), medical clinic
(kilometers), and marketplace (minutes by walking); at the village level. I check for whether the village has a ancestral hall or a clan with
population share >= 10%, the number of clans with population share >= 10%, whether it belongs to a minority area or a natural resource
area, and its distance to town center (kilometers) or county center (kilometers). Robust standard errors in parentheses are clustered at the
village/county level for columns 1–4/5–6. ∗∗∗ : significant at 1%; ∗∗ : significant at 5%; ∗ : significant at 10%.

22
W. Sha Journal of Development Economics 160 (2023) 102985

Table A.5
Dynamic effects of land expropriation on political outcomes.
Source: China Family Panel Studies.
Trust towards local officials Conflict with local officials
Unbalanced Balanced Unbalanced Balanced
(1) (2) (3) (4)
Panel A: Two-way fixed-effect
6 years before −0.0869 −0.0319 −0.0117 −0.0143
(0.1520) (0.1419) (0.0239) (0.0219)
4 years before −0.0174 0.0319 0.0035 0.0053
(0.1048) (0.1074) (0.0131) (0.0126)
0 −0.2382∗∗∗ −0.2442∗∗ 0.0238∗∗ 0.0262∗∗
(0.0896) (0.1022) (0.0106) (0.0114)
2 years after −0.1732 −0.1367 0.0247∗∗ 0.0292∗∗
(0.1114) (0.1209) (0.0117) (0.0119)
4 years after 0.0989 0.1113 0.0199 0.0291∗∗
(0.1679) (0.1698) (0.0143) (0.0144)
# of observations 78329 46400 73440 42827
Adj. R-squared 0.3239 0.3339 0.1403 0.1547
Trust towards local officials Conflict with local officials
Unbalanced Balanced Unbalanced Balanced
(1) (2) (3) (4)
Panel B: Sun and Abraham 2021
6 years before −0.1578 −0.0513 −0.0104 −0.0117
(0.1639) (0.1546) (0.0235) (0.0245)
4 years before −0.0473 0.0093 0.0043 0.0059
(0.1129) (0.1141) (0.0135) (0.0131)
0 −0.2767∗∗∗ −0.2802∗∗∗ 0.0276∗∗ 0.0267∗∗
(0.0925) (0.1045) (0.0109) (0.0118)
2 years after −0.2039∗ −0.1674 0.0214∗ 0.0226∗
(0.1181) (0.1319) (0.0126) (0.0130)
4 years after 0.0547 0.0869 0.0145 0.0317
(0.1736) (0.1924) (0.0248) (0.0228)
# of observations 78363 45828 73474 42425
Adj. R-squared 0.3240 0.3343 0.1406 0.1538
Individual fixed effects Yes Yes Yes Yes
Time fixed effects Yes Yes Yes Yes

Notes: Unit of observation is the individual-year. The sample is a four-year panel (2012, 2014, 2016, and
2018) for columns 1–2 and a four-year panel (2010, 2012, 2014, and 2016) for column 3–4. The sample
excludes individuals who had their land expropriated before 2012 (2010) for columns 1–2 (columns 3–4).
Trust towards local government officials is a categorical variable ranged from 0 (extremely low trust) to 10
(extremely high trust). Conflict with local government officials is a dummy that equals to 1 if the individual
had conflicts with local government officials. Robust standard errors in parentheses are clustered at the
village (of current residence) level. ∗∗∗ : significant at 1%; ∗∗ : significant at 5%; ∗ : significant at 10%.

23
W. Sha Journal of Development Economics 160 (2023) 102985

Table A.6
Dynamic effects of land expropriation on additional political outcomes.
Source: China Family Panel Studies.
Unfair treatment Unreasonable delay Evaluation of performance Perception of severity
by local officials at local agencies of county government of nationwide corruption
Unbalanced Balanced Unbalanced Balanced Unbalanced Balanced Unbalanced Balanced
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A: Two-way fixed-effect
8 years before −0.0159 −0.0569
(0.0672) (0.0679)
6 years before −0.0166 −0.0064 0.0067 0.0110 −0.0139 −0.0256 0.2043 0.2671
(0.0153) (0.0170) (0.0179) (0.0197) (0.0451) (0.0522) (0.1959) (0.2086)
4 years before −0.0131 −0.0025 −0.0017 0.0022 −0.0392 −0.0600 0.2140 0.2699
(0.0149) (0.0147) (0.0151) (0.0167) (0.0382) (0.0403) (0.1381) (0.1642)
0 0.0352∗∗∗ 0.0325∗∗ 0.0422∗∗∗ 0.0410∗∗∗ 0.0260 0.0200 0.2612∗∗ 0.2108∗
(0.0136) (0.0145) (0.0139) (0.0156) (0.0313) (0.0351) (0.1146) (0.1258)
2 years after 0.0154 0.0201 0.0296 0.0277 0.0391 0.0328 0.2634∗ 0.2142
(0.0183) (0.0180) (0.0193) (0.0219) (0.0435) (0.0422) (0.1468) (0.1795)
4 years after 0.0004 0.0017 0.0095 0.0090 0.0197 −0.0015 0.0261 −0.1115
(0.0200) (0.0210) (0.0240) (0.0261) (0.0463) (0.0485) (0.2076) (0.2324)
6 years after −0.0240 −0.0208
(0.0636) (0.0679)
# of villages 1501 842 1502 842 2047 1119 2030 1094
# of observations 73422 40634 72897 40387 92562 52142 70658 39808
Adj. R-squared 0.5158 0.4513 0.5276 0.4602 0.4802 0.4139 0.5220 0.4484
Unfair treatment Unreasonable delay Evaluation of performance Perception of severity
by local officials at local agencies of county government of nationwide corruption
Unbalanced Balanced Unbalanced Balanced Unbalanced Balanced Unbalanced Balanced
(1) (2) (3) (4) (5) (6) (7) (8)
Panel B: Sun and Abraham 2021
8 years before 0.0157 −0.0433
(0.0741) (0.0759)
6 years before −0.0475 −0.0436 −0.0239 −0.0200 0.0209 0.0012 0.0093 0.0158
(0.0304) (0.0318) (0.0364) (0.0365) (0.0585) (0.0650) (0.0744) (0.0758)
4 years before −0.0097 0.0014 0.0067 0.0149 −0.0436 −0.0472 0.0589 0.0888
(0.0207) (0.0215) (0.0200) (0.0213) (0.0463) (0.0492) (0.0550) (0.0563)
0 0.0321∗∗ 0.0284∗ 0.0432∗∗∗ 0.0461∗∗∗ 0.0239 0.0349 0.0849∗∗ 0.0841∗
(0.0152) (0.0170) (0.0155) (0.0161) (0.0372) (0.0437) (0.0427) (0.0441)
2 years after 0.0086 0.0163 0.0229 0.0257 0.0216 0.0211 0.1148∗ 0.1061
(0.0198) (0.0205) (0.0238) (0.0239) (0.0558) (0.0522) (0.0641) (0.0679)
4 years after 0.0032 0.0071 −0.0002 −0.0014 −0.0025 −0.0192 −0.0227 −0.0666
(0.0364) (0.0372) (0.0385) (0.0391) (0.0681) (0.0647) (0.0911) (0.0947)
6 years after −0.1829 −0.1501
(0.1168) (0.1160)
# of observations 73460 42386 72937 42099 92622 46466 70715 41334
Adj. R-squared 0.2003 0.2140 0.2178 0.2203 0.2225 0.2282 0.2197 0.2203
Individual fixed effects Yes Yes Yes Yes Yes Yes Yes Yes
Time fixed effects Yes Yes Yes Yes Yes Yes Yes Yes

Notes: Unit of observation is the individual-year. The sample is a four-year panel (2010, 2012, 2014, and 2016) for columns 1–4, a five-year panel
(2010, 2012, 2014, 2016, and 2018) for columns 5–6, and a four-year panel (2012, 2014, 2016, and 2018) for columns 7–8. Unfair treatment refers
to an individual’s experience of being unfairly treated by local government officials, which is a dummy variable that is 1 if the individual had such
experience in the past year and 0 otherwise. Unreasonable delay (stalling) refers to that when citizens come to local government agencies, government
officials will shirk responsibility rather than helping them (‘‘kicking balls’’), which is a dummy variable that is 1 if the individual had such experience
in the past year and 0 otherwise. Evaluation of performance refers to an individual’s evaluation of county government performance in the past year, which
is ranged from 1 to 5 (much worse, worse, the same, better, much better). Perception of corruption refers to an individual’s perception of the severity
of nationwide government corruption in China, which is ranged from 0 (extremely low) to 10 (extremely high). Robust standard errors in parentheses are
clustered at the village (of current residence) level. ∗∗∗ : significant at 1%; ∗∗ : significant at 5%; ∗ : significant at 10%.

24
W. Sha Journal of Development Economics 160 (2023) 102985

Table A.7
Dynamic effects of land expropriation on household outcomes.
Source: China Family Panel Studies.
Agricultural inomce Land assets At least one Salary income Total income Food consumption Saving assets
per capita per capita member migrated per capita per capita per capita per capita
(log) (log) (dummy) (log) (log) (log) (log)

Unbalanced Balanced Unbalanced Balanced Unbalanced Balanced Unbalanced Balanced Unbalanced Balanced Unbalanced Balanced Unbalanced Balanced
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)

Panel A: Two-way fixed-effect

6+ years before 0.0448 0.0953 −0.1961 −0.1300 0.0042 0.0186 −0.0881 −0.0795 −0.0235 −0.0352 0.0575 −0.0033 0.0182 0.0519
(0.1772) (0.1780) (0.1853) (0.1856) (0.0314) (0.0324) (0.2050) (0.2154) (0.0636) (0.0635) (0.0687) (0.0685) (0.1996) (0.2068)
4 years before 0.0218 0.0453 −0.1327 −0.1044 0.0064 0.0229 −0.0787 −0.0312 −0.0287 −0.0254 −0.0172 −0.0717 0.1106 0.1177
(0.1443) (0.1507) (0.1510) (0.1555) (0.0308) (0.0317) (0.1712) (0.1777) (0.0568) (0.0567) (0.0559) (0.0562) (0.1808) (0.1959)
0 −0.0482 −0.0991 −0.3688∗∗ −0.3998∗∗ 0.0582∗∗ 0.0777∗∗∗ 0.2282 0.3794∗∗ 0.0014 0.0352 0.0252 −0.0346 0.3211∗ 0.4094∗∗
(0.1307) (0.1427) (0.1633) (0.1776) (0.0263) (0.0286) (0.1537) (0.1655) (0.0537) (0.0551) (0.0501) (0.0510) (0.1776) (0.1929)
2 years after −0.2778∗ −0.2990∗∗ −0.2819 −0.3306∗ 0.0690∗∗ 0.0874∗∗∗ 0.3081∗ 0.3150∗ 0.0482 0.0613 −0.0084 −0.0394 0.1037 0.1683
(0.1450) (0.1513) (0.1839) (0.1872) (0.0319) (0.0326) (0.1725) (0.1723) (0.0573) (0.0561) (0.0528) (0.0539) (0.2187) (0.2236)
4 years after −0.3435∗ −0.2625 −0.4747∗∗ −0.3908∗ 0.0631∗ 0.0730∗∗ 0.1760 0.2342 −0.0555 −0.0282 0.0265 −0.0047 0.2042 0.2233
(0.1990) (0.1956) (0.2247) (0.2237) (0.0340) (0.0340) (0.1966) (0.1972) (0.0879) (0.0889) (0.0649) (0.0676) (0.2377) (0.2450)
6 years after −0.3431 −0.2352 −0.2861 −0.1763 0.0454 0.0544 0.1723 0.2163 0.1480 0.1489 0.1380∗ 0.1156 0.3222 0.3610
(0.2590) (0.2549) (0.3104) (0.3087) (0.0417) (0.0409) (0.2508) (0.2501) (0.1162) (0.1155) (0.0775) (0.0746) (0.3313) (0.3303)
# of observations 38997 27939 42184 30567 31735 22693 42220 30578 40394 29626 40959 29753 42211 30560
Adj. R-squared 0.5679 0.5338 0.5284 0.4687 0.2812 0.2856 0.3993 0.3817 0.3648 0.2971 0.3368 0.2819 0.3619 0.3510

Agricultural inomce Land assets At least one Salary income Total income Food consumption Saving assets
per capita per capita member migrated per capita per capita per capita per capita
(log) (log) (dummy) (log) (log) (log) (log)

Unbalanced Balanced Unbalanced Balanced Unbalanced Balanced Unbalanced Balanced Unbalanced Balanced Unbalanced Balanced Unbalanced Balanced
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)

Panel B: Sun and Abraham 2021

6 years before 0.0766 0.0227 0.5089 0.4464 0.0031 0.0179 −0.3898 −0.2479 −0.0008 0.0323 −0.2043 −0.1955 0.1657 −0.0309
(0.2563) (0.2780) (0.3176) (0.3018) (0.0781) (0.0828) (0.4191) (0.4182) (0.1382) (0.1399) (0.1551) (0.1616) (0.6327) (0.6663)
4 years before −0.1944 −0.2602 0.1350 0.0886 −0.0891 −0.1015 −0.4297 −0.5388∗ −0.0443 −0.0599 −0.1452 −0.1592 0.1395 0.1459
(0.2108) (0.2252) (0.3051) (0.3055) (0.0729) (0.0702) (0.3304) (0.3265) (0.1154) (0.1199) (0.1398) (0.1342) (0.3445) (0.3472)
0 −0.3810 −0.6126∗∗ −0.2890 −0.4285 −0.0656 −0.0641 −0.1612 −0.3485 0.1000 0.0502 0.1366 0.0873 0.1707 −0.0192
(0.2802) (0.3056) (0.2753) (0.2863) (0.0495) (0.0492) (0.2653) (0.2852) (0.1031) (0.1101) (0.0962) (0.1022) (0.3063) (0.3342)
2 years after −0.2015 −0.3450 −0.6937∗∗ −0.7885∗∗ −0.0059 0.0002 −0.1179 −0.1499 −0.0235 −0.0676 0.1131 0.0676 0.1112 0.0455
(0.3116) (0.3310) (0.3221) (0.3244) (0.0497) (0.0510) (0.3451) (0.3507) (0.1424) (0.1538) (0.1086) (0.1208) (0.3931) (0.4393)
4 years after −0.3932 −0.2679 −0.7593∗ −0.7076∗ 0.1027 0.1229 0.6228 0.4263 0.2211 0.1654 0.2501∗ 0.1791 0.3868 0.3380
(0.3605) (0.3736) (0.4005) (0.4068) (0.0774) (0.0777) (0.3906) (0.3902) (0.1359) (0.1365) (0.1497) (0.1683) (0.4367) (0.4639)
6 years after −0.6196 −0.6993 −1.0094∗ −0.8567 0.0180 −0.0083 0.3473 0.1608 0.1925 0.2547 0.1482 0.0710 0.1934 0.3133
(0.5363) (0.5762) (0.6010) (0.6023) (0.0709) (0.0779) (0.5253) (0.5946) (0.1855) (0.1841) (0.2196) (0.2558) (0.4991) (0.5773)
# of observations 34879 27909 37880 30574 27671 22616 37901 30588 36499 29626 36739 29780 37897 30570
Adj. R-squared 0.5390 0.5321 0.4758 0.4617 0.2808 0.2822 0.3831 0.3801 0.2973 0.2919 0.2782 0.2778 0.3473 0.3464
Individual fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Time fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Notes: Unit of observation is the household-year. The sample is a five-year panel (2010, 2012, 2014, 2016, and 2018). The sample excludes households whose land has been expropriated before 2010. Following Sun and Abraham (2021),
two periods are dropped in the regression in Panel B. Robust standard errors in parentheses are clustered at the village (of current residence) level. ∗∗∗ : significant at 1%; ∗∗ : significant at 5%; ∗ : significant at 10%.

Table A.8
Effects of land expropriation on political outcomes by governance quality.
Source: China Family Panel Studies.
Trust towards officials Conflict with officials Trust towards officials Conflict with officials Trust towards officials Conflict with officials
Bulletin info. available in 2014 Government transparency in 2018 Government corruption in 2018
No Yes No Yes Low High Low High High Low High Low
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
Panel A: Individual and time fixed effects
Land expropriation −0.6685∗∗∗ −0.1884 0.0675∗∗ 0.0393 −0.3634∗∗∗ 0.0210 0.0334∗∗ 0.0156 −0.2703∗ −0.0744 0.0369∗∗∗ 0.0133
(0.1638) (0.1717) (0.0327) (0.0440) (0.1288) (0.1289) (0.0164) (0.0124) (0.1405) (0.1255) (0.0143) (0.0141)
Adj. R-squared 0.6637 0.6560 0.7206 0.6933 0.5755 0.6065 0.4762 0.5048 0.6009 0.5763 0.4828 0.4947
𝐹 -test: Chi-sq. [p-value] 2.8293[0.0926] 0.1635[0.6859] 4.4627[0.0346] 0.7497[0.3866] 1.0809[0.2985] 1.3955[0.2375]
Panel B: Individual and village-by-year fixed effects
Land expropriation −0.6233∗∗∗ −0.0962 0.0518∗ 0.0123 −0.3041∗∗∗ 0.0120 0.0162 0.0039 −0.2833∗∗∗ −0.0265 0.0074 0.0114
(0.1660) (0.1192) (0.0300) (0.0196) (0.0903) (0.1131) (0.0122) (0.0103) (0.1070) (0.0979) (0.0109) (0.0114)
Adj. R-squared 0.6816 0.6747 0.7337 0.7086 0.6038 0.6337 0.5129 0.5494 0.6279 0.6050 0.5297 0.5294
𝐹 -test: Chi-sq. [p-value] 3.2781[0.0702] 0.4173[0.5183] 2.7397[0.0979] 0.3424[0.5584] 1.8000[0.1797] 0.0367[0.8480]
Land expro. mean 0.0674 0.1087 0.0537 0.0860 0.0576 0.0724 0.0594 0.0838 0.0624 0.0656 0.0653 0.0751
Land expro. SD 0.2507 0.3113 0.2255 0.2804 0.2330 0.2591 0.2363 0.2771 0.2419 0.2476 0.2471 0.2636
Dep. var. mean 5.3671 5.3049 0.0400 0.0442 5.2109 5.2039 0.0473 0.0487 5.1757 5.2381 0.0473 0.0485
Dep. var. SD 2.7260 2.7081 0.1959 0.2056 2.6762 2.6554 0.2122 0.2153 2.7323 2.6036 0.2122 0.2149
# of villages 150 252 150 251 576 576 461 457 577 575 461 457
# of observations 19272 31006 12605 19992 37675 29332 35334 28979 32551 34456 30945 33368

Notes: Unit of observation is the individual-year. The sample is a three-year panel (2014, 2016, and 2018) for columns 1–2, a two-year panel (2014 and 2016) for columns 3–4, a four-year panel (2012, 2014, 2016,
and 2018) for columns 5–6, 9–10, and 13–14, and a four-year panel (2010, 2012, 2014, and 2016) for columns 7–8, 11–12, and 15–16. Trust towards local government officials is a categorical variable ranged from 0
(extremely low trust) to 10 (extremely high trust). Conflict with local government officials is a dummy that equals to 1 if the individual had conflicts with local government officials in the past year for the 2012, 2014,
and 2016 waves, and equals to 1 if the individual has ever had conflicts with officials in history for the 2010 wave. Land expropriation is a dummy that is 1 if the individual is in a household whose land has ever been
expropriated in a given year. The sample is divided into subsamples based on: (i) whether the village reports that the village committee posts information about land expropriation projects on the bulletin board when
surveyed in 2014; (ii) the degree of government transparency at the city level; and (iii) the degree of government corruption at the city level. The null of the 𝐹 -test is there is no difference between the two coefficients
estimated from using different subsamples. Robust standard errors in parentheses are clustered at the village (of current residence) level. ∗∗∗ : significant at 1%; ∗∗ : significant at 5%; ∗ : significant at 10%.

25
W. Sha Journal of Development Economics 160 (2023) 102985

Table A.9
Effects of land expropriation on political outcomes: Governance quality.
Source: China Family Panel Studies.
Trust Conflict Trust Conflict
(1) (2) (3) (4)
Land expropriation −0.5059∗∗∗ 0.0430∗∗ −0.4453∗∗∗ 0.0142
(0.1623) (0.0191) (0.1179) (0.0139)
Land expro. X high transparency 0.4073∗∗ −0.0136 0.3227∗∗ −0.0123
(0.1759) (0.0199) (0.1427) (0.0159)
Land expro. X low corruption 0.2672 −0.0214 0.2642∗ 0.0036
(0.1761) (0.0195) (0.1424) (0.0156)
Land expro. + expro. X high trans. + expro. X low corrupt. 0.1686 0.0081 0.1417 0.0056
(0.1534) (0.0156) (0.1301) (0.0129)
# of villages 1150 917 1150 917
# of observations 67007 64313 67007 64313
Adj. R-squared 0.3264 0.1508 0.3418 0.1850
Individual fixed effects Yes Yes Yes Yes
Time fixed effects Yes Yes No No
Village FEs X Time FEs No No Yes Yes
Notes: Unit of observation is the individual-year. The sample is a four-year panel (2012, 2014, 2016, and 2018) for
columns 1 and 3, and a four-year panel (2010, 2012, 2014, and 2016) for columns 2 and 4. Trust towards local government
officials is a categorical variable ranged from 0 (extremely low trust) to 10 (extremely high trust). Conflict with local
government officials is a dummy that equals to 1 if the individual had conflicts with local government officials in the past
year for the 2012, 2014, and 2016 waves, and equals to 1 if the individual has ever had conflicts with officials in history
for the 2010 wave. Land expropriation is a dummy that is 1 if the individual is in a household whose land has ever been
expropriated in a given year. ‘‘high transparency’’ equals to 1 if the individual is in a city whose transparency level is
above the mean in 2018, and 0 otherwise; ‘‘low corruption’’ equals to 1 if the individual is in a city whose corruption
level is below the mean in 2018, and 0 otherwise. Robust standard errors in parentheses are clustered at the village (of
current residence) level. ∗∗∗ : significant at 1%; ∗∗ : significant at 5%; ∗ : significant at 10%.

Table A.10
Effects of land expropriation on political outcomes by project benefits.
Source: China Family Panel Studies.
Trust towards local officials Conflict with local officials
Village infrastructure improvement during 2010–2013
Yes No Yes No
(1) (2) (3) (4)
Panel A: Individual and time fixed effects
Land expropriation 0.0316 −0.5373∗ 0.0021 0.0023
(0.1687) (0.2818) (0.0209) (0.0224)
Adj. R-squared 0.3259 0.3645 0.1459 0.1688
𝐹 -test: Chi-sq. [p-value] 2.0218[0.1551] 0.0000[0.9963]
Panel B: Individual and village-by-year fixed effects
Land expropriation 0.0086 −0.4313∗∗∗ −0.0049 0.0061
(0.1799) (0.1513) (0.0198) (0.0155)
Adj. R-squared 0.3520 0.3806 0.1675 0.1800
𝐹 -test: Chi-sq. [p-value] 1.1526[0.2830] 0.0656[0.7979]
Land expro. mean 0.1102 0.1025 0.1105 0.1027
Land expro. SD 0.3131 0.3033 0.3135 0.3036
Dep. var. mean 5.3028 5.1967 0.0413 0.0387
Dep. var. SD 2.5830 2.6210 0.1990 0.1929
# of villages 206 196 206 196
# of observations 19962 18436 18887 17485
Notes: Unit of observation is the individual-year. The sample is a two-year panel (2012 and 2014) for columns 1–4. Trust towards local
officials is a categorical variable ranged from 0 (extremely low trust) to 10 (extremely high trust) in a given year. Conflict with local
officials is a dummy that equals to 1 if the individual had conflicts with local government officials in the past year. Land expropriation
is a dummy that is 1 if the individual is in a household whose land has ever been expropriated in a given year. The sample is divided
into subgroups based on whether the village has had access to at least one of the following infrastructure from January 1st, 2010 to
December 31st, 2013: electricity, cable radio, cable/satellite TV, postal service, telephone, cellphone signal/service, road, railway, tap
water, pipeline gas, public bus, and subway. The null of the 𝐹 -test is there is no difference between the two coefficients estimated from
using different subsamples. Robust standard errors in parentheses are clustered at the village (of current residence) level. ∗∗∗ : significant
at 1%; ∗∗ : significant at 5%; ∗ : significant at 10%.

26
W. Sha Journal of Development Economics 160 (2023) 102985

Table A.11
Effects of land expropriation on political outcomes by agricultural dependence.
Source: China Family Panel Studies.
Trust towards local officials Conflict with local officials
Non-agricultural employment in 2010
No Yes No Yes
(1) (2) (3) (4)
Panel A: Individual and time fixed effects
Land expropriation −0.3460∗∗∗ −0.1517 0.0420∗∗∗ 0.0131
(0.1284) (0.1549) (0.0163) (0.0138)
Adj. R-squared 0.3016 0.3584 0.1335 0.1780
𝐹 -test: Chi-sq. [p-value] 1.0071[0.3156] 2.4161[0.1201]
Panel B: Individual and village-by-year fixed effects
Land expropriation −0.3183∗∗∗ −0.2137 0.0365∗∗∗ −0.0094
(0.1111) (0.1513) (0.0130) (0.0126)
Adj. R-squared 0.3269 0.3814 0.1873 0.2098
𝐹 -test: Chi-sq. [p-value] 1.0071[0.3156] 4.0997[0.0429]
Land expro. mean 0.0659 0.0754 0.0739 0.0895
Land expro. SD 0.2482 0.2641 0.2616 0.2855
Dep. var. mean 5.4764 5.1271 0.0505 0.0451
Dep. var. SD 2.6438 2.5994 0.2190 0.2076
# of villages 519 551 465 511
# of observations 27751 18437 26163 18569
Notes: Unit of observation is the individual-year. The sample is a four-year panel (2012, 2014, 2016, and 2018) for columns 1–2, and a four-year panel (2010,
2012, 2014, and 2016) for columns 3–4. Trust towards local officials is a categorical variable ranged from 0 (extremely low trust) to 10 (extremely high trust)
in a given year. Conflict with local officials is a dummy that equals to 1 if the individual had conflicts with local government officials in the past year. Land
expropriation is a dummy that is 1 if the individual is in a household whose land has ever been expropriated in a given year. The sample is divided into subgroups
based on whether the household has at least one member working in the non-agricultural sector in the baseline year 2010. The null of the 𝐹 -test is there is no
difference between the two coefficients estimated from using different subsamples. Robust standard errors in parentheses are clustered at the village (of current
residence) level. ∗∗∗ : significant at 1%; ∗∗ : significant at 5%; ∗ : significant at 10%.

Fig. A.1. Annual land conversion income at the national level. Note: The black squares plot annual land conversion income; the monetary unit is one trillion yuan.
Source: Ministry of Finance of China.

27
W. Sha Journal of Development Economics 160 (2023) 102985

Fig. A.2. The time trend of political outcomes by group. Notes: The figure plots the time evolution of political outcomes (mean of all the individual-year observations) by group.
In Panels A and B, red line is the time trend of the outcome for rural non-expropriated group, blue line for rural expropriated group, green line for urban non-expropriated group,
and pink line for urban expropriated group. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Source: China Family Panel Studies.

Fig. A.3. CFPS sample attrition rate. Notes: Blue circles indicate the attrition rate between every two consecutive years for the constructed sample, red triangles for the expropriated
sample, green rectangles for the non-expropriated sample.
Source: China Family Panel Studies.

28
W. Sha Journal of Development Economics 160 (2023) 102985

Fig. A.4. Dynamic effects of land expropriation on additional political outcomes. Notes: Unit of observation is the individual-year. The sample is a four-year panel (2010, 2012,
2014, and 2016) for Panels A and B, a five-year panel (2010, 2012, 2014, 2016, and 2018) for Panel C, and a four-year panel (2012, 2014, 2016, and 2018) for Panel D. The
sample excludes always-expropriated individuals. The shaded region indicates the expropriation was happening during this time period. The CFPS survey is conducted in every
two years, thus, the gap between two consecutive periods is two years (the horizontal axis).
Source: China Family Panel Studies.

Fig. A.5. Dynamic effects of land expropriation on additional political outcomes. Notes: Unit of observation is the individual-year. The sample is a four-year panel (2010, 2012,
2014, and 2016) for Panels A and B, a five-year panel (2010, 2012, 2014, 2016, and 2018) for Panel C, and a four-year panel (2012, 2014, 2016, and 2018) for Panel D. The
sample excludes always-expropriated individuals. The shaded region indicates the expropriation was happening during this time period. The CFPS survey is conducted in every
two years, thus, the gap between two consecutive periods is two years (the horizontal axis).
Source: China Family Panel Studies.

29
W. Sha Journal of Development Economics 160 (2023) 102985

Fig. A.6. Dynamic effects of land expropriation on household outcomes. Notes: Unit of observation is the household-year. The sample is a five-year household panel (2010, 2012,
2014, 2016, and 2018) and excludes households who had their land expropriated before 2010. The shaded region indicates the expropriation was happening during this time
period. The CFPS survey is conducted in every two years, thus, the gap between two consecutive periods is two years (the horizontal axis).
Source: China Family Panel Studies.

30
W. Sha Journal of Development Economics 160 (2023) 102985

References Dahlberg, S., Linde, J., 2018. Socialization or experience? Institutional trust and
satisfaction with democracy among emigrants in different institutional settings. J.
Acemoglu, D., Johnson, S., Robinson, J.A., 2001. The colonial origins of comparative Politics 80 (4), 1389–1393. http://dx.doi.org/10.1086/698661.
development: An empirical investigation. Amer. Econ. Rev. 91 (5), 1369–1401. de Janvry, A., Emerick, K., Gonzalez-Navarro, M., Sadoulet, E., 2015. Delinking land
http://dx.doi.org/10.1257/aer.91.5.1369. rights from land use: Certification and migration in Mexico. Amer. Econ. Rev. 105
Acemoglu, D., Johnson, S., Robinson, J.A., 2002. Reversal of fortune: Geography and (10), 3125–3149. http://dx.doi.org/10.1257/aer.20130853.
institutions in the making of the modern world income distribution. Q. J. Econ. Goldstein, M., Houngbedji, K., Kondylis, F., O’Sullivan, M., Selod, H., 2018. For-
117 (4), 1231–1294. http://dx.doi.org/10.1162/003355302320935025. malization without certification? Experimental evidence on property rights and
Acemoglu, D., Johnson, S., Robinson, J.A., 2005. Institutions as a fundamental cause investment. J. Dev. Econ. 132, 57–74. http://dx.doi.org/10.1016/j.jdeveco.2017.
of long-run growth. In: Aghion, P., Durlauf, S.N. (Eds.), In: Handbook of Economic 12.008.
Growth, vol. 1A, Elsevier, Oxford, pp. 385–472. http://dx.doi.org/10.1016/S1574- Goldstein, M., Udry, C., 2008. The profits of power: Land rights and agricultural
0684(05)01006-3, chapter 6. investment in Ghana. J. Polit. Econ. 116 (6), 981–1022. http://dx.doi.org/10.1086/
Acemoglu, D., Robinson, J.A., 2012. Why Nations Fail: The Origins of Power, Prosperity, 595561.
and Poverty. Crown Books. Guo, X., 2001. Land expropriation and rural conflicts in China. China Q. 166, 422–439.
Acemoglu, D., Robinson, J.A., 2019. Rents and economic development: The perspective http://dx.doi.org/10.1017/S0009443901000201.
of why nations fail. Public Choice 181 (1–2), 13–28. http://dx.doi.org/10.1007/ Hall, R.E., Jones, C.I., 1999. Why do some countries produce so much more output
s11127-019-00645-z. per worker than others? Q. J. Econ. 114 (1), 83–116. http://dx.doi.org/10.1162/
Alston, L.J., Libecap, G.D., Mueller, B., 1998. Property rights and land conflict: A 003355399555954.
comparison of settlement of the U.S. Western and Brazilian Amazon frontiers. In: Han, L., Kung, J.K.S., 2015. Fiscal incentives and policy choices of local governments:
Coatsworth, J.H., Taylor, A.M. (Eds.), Latin America and the World Economy Since Evidence from China. J. Dev. Econ. 116, 89–104. http://dx.doi.org/10.1016/j.
1800. Harvard University Press, Cambridge, MA. jdeveco.2015.04.003.
Anseeuw, W., Boche, M., Breu, T., Giger, M., Lay, J., Messerli, P., Nolte, K., 2012. He, G., Wang, S., 2017. Do college graduates serving as village officials help rural
Transnational land deals for agriculture in the global south: Analytical report based China? Am. Econ. J.: Appl. Econ. 9 (4), 186–215. http://dx.doi.org/10.1257/app.
on the land matrix database. 20160079.
Banerjee, A., Duflo, E., Qian, N., 2020. On the road: Access to transportation He, G., Xie, Y., Zhang, B., 2020. Expressways, GDP, and the environment: The case
infrastructure and economic growth in China. J. Dev. Econ. 145, 102442. http: of China. J. Dev. Econ. 145, 102485. http://dx.doi.org/10.1016/j.jdeveco.2020.
//dx.doi.org/10.1016/j.jdeveco.2020.102442. 102485.
Banerjee, A.V., Gertler, P.J., Ghatak, M., 2002. Empowerment and efficiency: Tenancy He, S., Xue, D., 2014. Identity building and communal resistance against landgrabs in
reform in West Bengal. J. Polit. Econ. 110 (2), 239–280. http://dx.doi.org/10.1086/ Wukan village, China. Curr. Anthropol. 55 (S9), S126–S137. http://dx.doi.org/10.
338744. 1086/676132.
Bardhan, P., 2012. Awakening Giants, Feet of Clay: Assessing the Economic Rise of Hidalgo, F.D., Naidu, S., Nichter, S., Richardson, N., 2010. Economic determinants of
China and India. Princeton University Press. land invasions. Rev. Econ. Stat. 92 (3), 505–523.
Bardhan, P., 2016. State and development: The need for a reappraisal of the current Huet-Vaughn, E., 2019. Stimulating the vote: ARRA road spending and vote share. Am.
literature. J. Econ. Lit. 54 (3), 862–892. http://dx.doi.org/10.1257/jel.20151239. Econ. J.: Econ. Policy 11 (1), 292–316. http://dx.doi.org/10.1257/pol.20170151.
Bardhan, P., 2020. The Chinese governance system: Its strengths and weaknesses Jacoby, H.G., Li, G., Rozelle, S., 2002. Hazards of expropriation: Tenure insecurity and
in a comparative development perspective. China Econ. Rev. 61, 101430. http: investment in rural China. Amer. Econ. Rev. 92 (5), 1420–1447. http://dx.doi.org/
//dx.doi.org/10.1016/j.chieco.2020.101430. 10.1257/000282802762024575.
Bargain, O., Aminjonov, U., 2020. Trust and compliance to public health policies Johnson, S., McMillan, J., Woodruff, C., 2002. Property rights and finance. Amer. Econ.
in times of COVID-19. J. Public Econ. 192, 104316. http://dx.doi.org/10.1016/ Rev. 92 (5), 1335–1356. http://dx.doi.org/10.1257/000282802762024539.
j.jpubeco.2020.104316. Khwaja, A.I., 2009. Can good projects succeed in bad communities? J. Public Econ. 93
Besley, T., 1995. Property rights and investment incentives: Theory and evidence from (7–8), 899–916. http://dx.doi.org/10.1016/j.jpubeco.2009.02.010.
Ghana. J. Polit. Econ. 103 (5), 903–937. http://dx.doi.org/10.1086/262008. Landesa, 2012. Findings from Landesa’s survey of rurul China published. Available
Besley, T., Ghatak, M., 2010. Property rights and economic development. In: Rodrik, D., online: https://www.landesa.org/press-and-media/6th-china-survey/. (Accessed 24
Rosenzweig, M. (Eds.), In: Handbook of Development Economics, vol. 5, Elsevier, June 2019).
Oxford, pp. 4525–4595. http://dx.doi.org/10.1016/B978-0-444-52944-2.00006-9, Lewis-Faupel, S., Neggers, Y., Olken, B.A., Pande, R., 2016. Can electronic procure-
chapter 68. ment improve infrastructure provision? Evidence from public works in India and
Besley, T., Persson, T., 2011. Pillars of Prosperity: The Political Economics of Indonesia. Am. Econ. J.: Econ. Policy 8 (3), 258–283. http://dx.doi.org/10.1257/
Development Clusters. Princeton University Press. pol.20140258.
Björkman, M., Svensson, J., 2009. Power to the people: Evidence from a randomized Li, H., Zhou, L.A., 2005. Political turnover and economic performance: The incentive
field experiment on community-based monitoring in Uganda. Q. J. Econ. 124 (2), role of personnel control in China. J. Public Econ. 89 (9–10), 1743–1762. http:
735–769. http://dx.doi.org/10.1162/qjec.2009.124.2.735. //dx.doi.org/10.1016/j.jpubeco.2004.06.009.
Borusyak, K., Jaravel, X., Spiess, J., 2022. Revisiting event study designs: Robust and Lin, J.Y., 1992. Rural reforms and agricultural growth in China. Amer. Econ. Rev. 82
efficient estimation. Working Paper. (1), 34–51.
Cai, F., Park, A., Zhao, Y., 2008. The Chinese labor market in the reform era. In: Lü, X., 2014. Social policy and regime legitimacy: The effects of education reform
Brandt, L., Rawski, T.G. (Eds.), China’s Great Economic Transformation. Cambridge in China. Am. Political Sci. Rev. 108 (2), 423–437. http://dx.doi.org/10.1017/
University Press, pp. 167–214. http://dx.doi.org/10.1017/CBO9780511754234. S0003055414000124.
Cantoni, D., Chen, Y., Yang, D.Y., Yuchtman, N., Zhang, Y.J., 2017. Curriculum and Lu, Y., Wang, J., Zhu, L., 2019. Place-based policies, creation, and agglomeration
ideology. J. Polit. Econ. 125 (2), 338–392. http://dx.doi.org/10.1086/690951. economies: Evidence from China’s economic zone program. Am. Econ. J.: Econ.
Chari, A., Liu, E.M., Wang, S.Y., Wang, Y., 2022. Property rights, land misallocation Policy 11 (3), 325–360. http://dx.doi.org/10.1257/pol.20160272.
and agricultural efficiency in China. Rev. Econom. Stud. 88 (4), 1831–1862. Ma, S., Mu, R., 2020. Forced off the farm? Farmers’ labor allocation response to
http://dx.doi.org/10.1093/restud/rdaa072. land requisition in China. World Dev. 132, 104980. http://dx.doi.org/10.1016/
Chen, L., 2014. Legal and institutional analysis of land expropriation in China. In: j.worlddev.2020.104980.
Fu, H., Gillespie, J. (Eds.), Resolving Land Disputes in East Asia: Exploring the Manacorda, M., Miguel, E., Vigorito, A., 2011. Government transfers and political
Limits of Law. Cambridge University Press, pp. 59–85. http://dx.doi.org/10.1017/ support. Am. Econ. J.: Appl. Econ. 3 (3), 1–28. http://dx.doi.org/10.1257/app.
CBO9781107589193.006. 3.3.1.
Chen, T., Kung, J.K.S., 2019. Busting the ‘‘princelings’’: The campaign against cor- Mattingly, D.C., 2016. Elite capture: How decentralization and informal institutions
ruption in China’s primary land market. Q. J. Econ. 134 (1), 185–226. http: weaken property rights in China. World Politics 68 (3), 383–412. http://dx.doi.
//dx.doi.org/10.1093/qje/qjy027. org/10.1017/S0043887116000083.
Chen, Y., Yang, D.Y., 2019a. Historical Traumas and the Roots of Political Distrust: Mattingly, D.C., 2019. The Art of Political Control in China. Cambridge University
Political Inference from the Great Chinese Famine. Working Paper. Press, http://dx.doi.org/10.1017/9781108662536.
Chen, Y., Yang, D.Y., 2019b. The impact of media censorship: 1984 or brave new Newton, K., Stolle, D., Zmerli, S., 2018. Social and political trust. In: Uslander, E.M.
world? Amer. Econ. Rev. 109 (6), 2294–2332. http://dx.doi.org/10.1257/aer. (Ed.), The Oxford Handbook of Social and Political Trust. Oxford Univer-
20171765. sity Press Oxford, New York, pp. 961–976. http://dx.doi.org/10.1093/oxfordhb/
Chen, D.L., Yeh, S., 2020. Growth Under the Shadow of Expropriation? The Economic 9780199270125.003.0018, chapter 3.
Impacts of Eminent Domain. Working Paper. Nie, H., Han, G., Ma, L., Zhang, N., 2019. The ranking of government-business
Cui, E., Tao, R., Warner, T.J., Yang, D.L., 2015. How do land takings affect political relationship among Chinese cities (2018) [Zhongguo Chengshi Zhengshang Gaunxi
trust in rural China? Political Stud. 63, 91–109. http://dx.doi.org/10.1111/1467- Paihangbang (2018)]. Renmin University of China Research Report. (in Chinese).
9248.12151. North, D.C., 1981. Structure and Change in Economic History. Norton.
Cull, R., Xu, L.C., 2005. Institutions, ownership, and finance: The determinants of North, D.C., Weingast, B.R., 1989. Constitutions and commitment: The evolution
profit reinvestment among Chinese firms. J. Financ. Econ. 77 (1), 117–146. http: of institutions governing public choice in seventeenth-century England. J. Econ.
//dx.doi.org/10.1016/j.jfineco.2004.05.010. History 49 (4), 803–832. http://dx.doi.org/10.1017/S0022050700009451.

31
W. Sha Journal of Development Economics 160 (2023) 102985

Olken, B.A., 2007. Monitoring corruption: Evidence from a field experiment in Wang, X., Fan, G., Hu, L., 2019. Marketization Index of China’s Provinces (2018)
Indonesia. J. Polit. Econ. 115 (2), 200–249. http://dx.doi.org/10.1086/517935. [Zhongguo Fensheng Shichanghua Zhishu Baogao (2018)]. China Social Sciences
Page, L., Pande, R., 2018. Ending global poverty: Why money isn’t enough. J. Econ. Academic Press, (in Chinese).
Perspect. 32 (4), 173–200. http://dx.doi.org/10.1257/jep.32.4.173. Wang, Z., Zhang, Q., Zhou, L.A., 2020. Career incentives of city leaders and urban
Park, A., Wang, S., 2010. Community-based development and poverty alleviation: An spatial expansion in China. Rev. Econ. Stat. 102 (5), 897–911. http://dx.doi.org/
evaluation of China’s poor village investment program. J. Public Econ. 94 (9–10), 10.1162/rest_a_00862.
790–799. http://dx.doi.org/10.1016/j.jpubeco.2010.06.005. Wong, H.L., Wang, Y., Luo, R., Zhang, L., Rozelle, S., 2017. Local governance and the
Reinikka, R., Svensson, J., 2004. Local capture: Evidence from a central government quality of local infrastructure: Evidence from village road projects in rural China.
transfer program in Uganda. Q. J. Econ. 119 (2), 679–705. http://dx.doi.org/10. J. Public Econ. 152, 119–132. http://dx.doi.org/10.1016/j.jpubeco.2017.06.006.
1162/0033553041382120. Xiong, W., 2018. The Mandarin Model of Growth. National Bureau of Economic
Scott, J.C., 1985. Weapons of the Weak: Everyday Forms of Peasant Resistance. Yale Research Working Paper, http://dx.doi.org/10.3386/w25296.
University Press. Xu, C., 2011. The fundamental institutions of China’s reforms and development. J.
Sha, W., Zou, X., 2022. Can Eminent Domain Deliver Benefits to Households? Evidence Econ. Lit. 49 (4), 1076–1151. http://dx.doi.org/10.1257/jel.49.4.1076.
from China. Working Paper. Xu, Y., Yao, Y., 2015. Informal institutions, collective action, and public investment in
Sun, L., Abraham, S., 2021. Estimating dynamic treatment effects in event studies rural China. Am. Political Sci. Rev. 109 (2), 371–391. http://dx.doi.org/10.1017/
with heterogeneous treatment effects. J. Econometrics 225 (2), 175–199. http: S0003055415000155.
//dx.doi.org/10.1016/j.jeconom.2020.09.006. Zhang, Q., Feng, L., 2015. China’s land expropriation institutions: Process, problems,
Tilly, C., 1990. Coercion, Capital, and European States, AD 990-1992. Blackwell. and solutions [Zhongguo Zhengdi Zhidu: Guocheng, Kunjing Yu Chulu]. Renmin
van der Meer, T.W.G., 2018. Economic performance and political trust. In: Uslan- University of China Research Report. (in Chinese).
der, E.M. (Ed.), The Oxford Handbook of Social and Political Trust. Oxfrod Zhao, X., 2020. Land and labor allocation under communal tenure: Theory and evidence
University Press, New York, pp. 599–616. http://dx.doi.org/10.1093/oxfordhb/ from China. J. Dev. Econ. 147, 102526. http://dx.doi.org/10.1016/j.jdeveco.2020.
9780190274801.013.16, chapter 25. 102526.

32