Content uploaded by Christian Goebel
Author content
All content in this area was uploaded by Christian Goebel on Oct 15, 2017
Content may be subject to copyright.
Regulations Against Revolution:
Mapping Policy Innovations in China
Xuelian Chen
China Center for Global Governance and Development Strategies
chenxuelian@hotmail.com
Christian Göbel (corresponding author)*
University of Vienna, Department of East Asian Studies
Christian.Goebel@univie.ac.at
Journal of Chinese Governance (forthcoming 2016)
January 25, 2016
Abstract
Since 2004, the central government has incentivized local ocials to
solve regional problems proactively and to prevent their emergence. The
present contribution investigates the impact of these incentives. It ana-
lyzes the geographic distribution of policy innovations in China, the types
of measures local government ocials devise to solve or to prevent local
problems, and the factors to render likely the successful implementation
of certain measures at specic locations. Based on the literature on policy
implementation in China’s local states, we formulate hypotheses and test
them by means of geographically visualized data and multivariate linear
regressions. Our data source is composed of roughly 1,030 descriptions of
policy innovations implemented by local governments. These documents
have been submitted to compete for a renowned prize rewarding local pol-
icy innovation, and the existence of the measures described therein have
been veried by the organizers of the competition. Overall, the results of
our investigation suggest that the central government’s alteration of local
incentive structures exert a broad eect, thus providing the prerequisites
for the comprehensive inuence of local innovations on the entire system.
Whether they actually increase the legitimacy of the regime must however
be determined in a follow-up study.
*The research for this contribution was generously funded by the Swedish Science Council
(Vetenskapsrådet Projects No. 2011-1495). We would like to thank Yu Junbo for his help in
coding the documents.
1
Introduction
According to representative opinion surveys, the majority of the Chinese pop-
ulation is favorably disposed towards the authoritarian one-party regime it is
governed by (Kennedy 2009). At rst glance, this result seems dicult to rec-
oncile with the steady increase of social protests in China in recent years (Göbel
& Ong 2012). It seems equally counterintuitive that a non-democratic regime
should receive support rates of more than 70 percent of those surveyed (Kennedy
2009).
A convincing, commonplace explanation for this phenomenon is that such opin-
ion surveys typically measure the population’s general attitude towards “the
government”, without, however, taking into account that the Chinese polity
consists of ve administrative levels and is governed by two interwoven organi-
sations: the Chinese Communist Party (CCP) and the government. Typically,
the respondents associate “the government” with the CCP leadership in Beijing,
which takes great care to display concern about the livelihood of the people,
and which can easily blame shortcomings in policy implementation on subordi-
nate administrations (So 2013). The population’s judgement of local ocials
is believed to be far less benign, which goes far towards explaining how high
condence in the government can coexist with massive social unrest.
A large number of case studies detailing how the (rural) population suered at
the hand of greedy, corrupt and inept local ocials lent credible evidence to
the assertion that people harbour favourable attitudes towards the center, but
were critical of local ocials. While ocial documents released by the central
government ritually chided local leaders for not properly implementing manda-
tory social policies, scholars highlighted the negative consequences of “selective
policy implementation”, “implementation gaps” and the general challenge of im-
plementing policies that did not yield economic rents. Hence, in a rare display
of unity, Chinese scholars, non-Chinese scholars and the central government
alike identied local administrations as the bottleneck of China’s sustainable
development.
However, this consensus might no longer hold true. In a landmark document
passed in 2004, the central government identied the decits in local gover-
nance as one of the main factors that could potentially cause China’s one-party
regime to collapse. As a consequence, local leaders were subjected to immense
pressure by the central government to improve local governance. In what Chris-
tian Goebel and Thomas Heberer have referred to as a “policy innovation im-
perative”, the careers of local ocials became increasingly dependent on the
production of local policy innovations. 1
1The term “policy innovation” refers to “the development of new ideas or concepts and the
conversion of these ideas and concepts into new policies or policy instruments” (Göbel,
Heberer 2016).‘ Policy innovations are dierent from policy experiments, which are typically
devised on a higher administrative level and implemented locally by way of trial (Heilmann
2008b). They are also dierent from reforms, a term that refers to the alteration of existing
2
Since these measures have remained under-appreciated by scholars of Chinese
local politics, it is not surprising that their outcomes remain equally opaque.
Although recent studies do highlight the existence of local policy innovations,
the latter are not attributed to the central government’s policy innovation im-
perative. Hence, the consensus that local governments are characterised by low
skill levels and a lack of responsiveness to social demands is being challenged
by evidence to the contrary. However, the comparatively low number of case
studies employed by either side makes it fairly easy to accuse the other side of
selecting only those cases that support their respective position. The situation
becomes even more confused when ndings from the 1990s are compared with
research from the 2010s without, however, investigating processes of adaptation
and learning that might have taken place in between.2In more concise terms:
selection bias blinds us to variation, and the neglect of the policy innovation
imperative makes us impervious to the possibility of change.
The present study addresses both issues by analysing the impact of the policy
innovation imperative on the generation of policy innovations in China’s cities
and counties while accounting for local variation at the same time. To achieve
this, we investigate a large sample of policy innovations that we believe to be
representative for all policy innovations in China. Our data consists of 1,024
descriptions of policy innovations implemented by local governments. These
documents have been submitted to compete for a renowned prize rewarding
local policy innovation, and the existence of the measures described therein
have been veried by the organisers of the competition.
The analysis conrms that the number of local policy innovations has indeed
seen an abrupt rise after 2004, reveals that these innovations span a broad spec-
trum of measures, and that geographical location and the level of economic
development barely inuence the probability of whether a measure will be ini-
tiated or not. Overall, the results of our investigation suggest that the central
government’s alteration of local incentive structures exert a broad eect across
China, the notable exceptions being Tibet, Xinjiang, Gansu, and Inner Mongo-
lia.
Policy innovation on the local level
Scholars who studied rural China in the 1990s and 2000s will remember three
monographs. The rst, Li Changping’s I Spoke the Truth to the Premier (wo
xiang zongli shuo shihua), exposed the corruption and mismanagement by local
ocials in a county in Hubei (Li 2002). To no small degree, the book was
policies.‘
2This is a mistake almost made by the corresponding author in Göbel/Heberer 2016, and
he is grateful to Vivienne Shue and Patricia Thornton for alerting him to this error. The
attention to sequencing in the present contribution is the result of the invaluable exchanges
with these eminent scholars.
3
considered a credible source because the author had himself been a local ocial
since the early 1980s. The second monograph is An Investigative Report on
the Chinese Peasantry (Zhongguo nongmin diaocha) by journalists Chen Guidi
and Wu Chuntao (Chen & Wu 2004). In the rst part of the book, the authors
analyze in great detail what can euphemistically be called “strained relations
between peasants and cadres” in the North of Anhui Province. Given the graphic
detail with which the authors describe the extreme brutality applied by local
ocials and their henchmen in the process of collecting taxes, the book became
a bestseller despite being banned immediately after its publication. The third
book is China Along the Yellow River (Cao 2000), another bestseller in China
that described the craftiness with which local ocials engaged in corruption and
their utter disregard of the welfare of the people subjected to their rule.
The number of articles and monographs devoted to exposing the corruption,
lack of skill and utter dysfunctionality of local government published inside and
outside China is too high to do justice to even the most inuential works, but
all added to the broad consensus that the quality of local government in China
was dismal.3Broad agreement existed that local ocials invested energy only
in those policies that promised them economic rents, but had to be coerced into
implementing welfare-related policies (Heilmann 2008b; Hillman 2010; Smith
2009; Smith 2010).
The central government made no attempt to counter this impression. To the
contrary, they reinforced the paradigm of incapable local ocials by ritually
blaming them, in the preamble of rural reform documents, for the low quality of
public services and the existence of various other local problems. It is well known
that many of the problems local ocials had been blamed for were of a structural
nature and therefore the responsibility of the central government, but the latter
largely managed to stay clear of such accusations (Zhong 2015). In a notable
study published in 2004, the political scientist Lianjiang Li tested the theory
that the Chinese population trusted the central government more than its local
counterparts. His representative survey indeed found that the population trusts
those politicians less who hold a lower rank in the administrative hierarchy (Li
2004, see also Li 2008).
More recent publications present the mirror image of the paradigmatic inept lo-
cal ocial. Arguably inspired by Sebastian Heilmanns inuential publications
on policy experimentation in China (Heilmann 2008a; Heilmann 2008b; Heil-
mann 2009; Heilmann & Perry 2011), scholars began describing locations where
the quality of local governance was considerably better than in the places de-
scribed in the aforementioned studies (Chen & Yang 2009; Florini et al 2012;
Foster 2004; Teets & Hurst 2014; Zhu 2012; Zhu 2014). However, none of the
3A partial exception is Chen and Chun’s work quoted above, which, after describing the
horrible fate of people who refused to pay their taxes, highlight several local ocials who tried
to improve the lot of the people. Unfortunately, the rst part of the book received far more
attention than the second part, which, for example, is not included in the English translation
of the book (Chen & Wu 2007).
4
studies has sought to reconcile these rather uplifting ndings with the bleak
egy constructed by the previous wave of scholarship.
Faced with two seemingly contradictory visions of governance in local China, the
task of China scholars should be to nd explanations for the variation instead
of trying to essentialize China on the basis of very few case studies (Göbel
2011). Hence, the purpose of this article is to examine the variation of policy
innovations not only across space, but also across time. Besides allowing for the
possibility that some localities perform better than other locations in designing
and implementing policy innovations, we also account for processes of change
in one and the same location.
This is especially expedient as the central government has since 2004 attempted
to alter the incentive structure faced by local ocials in a way that discouraged
corruption and abuse of oce while rewarding policy innovations. Borne out of
the CCP party elites’ systematic analysis of the causes of the breakdown of the
Soviet Union4, they issued the “Decision to Strengthening the Governance Ca-
pacity of the Chinese Communist Party” (hereforth: Decision). With unusual
frankness, the Decision stated that the CCP was not immune to a Soviet-style
breakdown. A close reading of the Decision and subsequent policy documents
illustrates that central party elites hold the Soviet Union’s lack of responsive-
ness to the concerns of the population responsible for the demise of the Soviet
empire. After the Decision was passed, local politicians have increasingly been
pushed towards proving their ability to proactively solve local problems in ac-
cordance with the preferences of the central government. Göbel and Heberer
refer to this as the “policy innovation imperative” and, in their eld studies,
have conrmed that the altered incentives have indeed changed the political
attitudes and behaviour of local leaders ( Göbel & Heberer 2016).
We set out to prove that that these measures have had a systemic impact. More
specically, we believe that they have led to a an increase in policy innovations
across China. However, we also believe that owing to China’s considerable
heterogeneity, this impact is not uniform and will vary as the result of political
and structural factors. Our analysis will be guided by the following questions:
• The rst fundamental question is whether the incentives from the central
government have a systemic impact, or whether the local policy innova-
tions mentioned above constitute deviant cases. In the former case, pol-
icy innovations should mostly appear after 2004, when the Decision to
Strengthen the Governing Capacity of the CPC was announced.
• Second, given that the chief purpose of the Decision was reducing and pre-
venting social unrest by through higher responsiveness to social demands,
we expect policy innovations aimed at social welfare improvements to be
more prominent than those solely concerned with economic growth.
4David Shambaugh devotes a seizable part of his monograph to discussing this process
Shambaugh 2008.
5
• Third, we assume that especially those places lacking the administrative
infrastructure for designing and implementing social welfare policies will
have engaged in administrative reforms. This especially applies to poorer
regions.
• Fourth, responsiveness to popular demands does not only necessitate a
well-functioning administrative infrastructure, but also instruments to ag-
gregate popular demands. Hence, we expect policy innovations designed
to improve the communication between state and the society to also be
prevalent.
• Fifth, as regards the geographical distribution of policy innovations, we
expect countervailing pressures to be at work. On the one hand, since the
careers of local leaders were tied to the production of policy innovations,
we expect such innovations to be spread evenly across China. On the
other hand, the realisation of policy innovations is often tied to high costs,
especially where social welfare is concerned. Hence, we expect welfare-
related policy innovations to be especially prominent in the richer areas
of China, which are mainly found in the east coast provinces.
• Finally, as local municipalities compete against each other, we anticipate
contagion eects. In order to not fall behind the pioneers in one locality, a
policy innovation in one location should be followed by policy innovations
within the superordinate administrative unit.
Methods and data
Identifying the universe of cases is the major challenge confronting those who
wish to analyze the emergence and impacts of local policy innovations. To meet
this challenge, we make use of a database of local government innovations that
contains detailed and useful information on the emergence and implementation
of more than 1,000 locally initiated policies. More specically, the database
consists of 1,024 applications for participation in a “Government Innovation
Award” presented by the China Centre for Comparative Politics and Economics
(CCCPE). The CCCPE is a think tank belonging to the CCP Central Commit-
tee’s Bureau for Translation and Editing. Its main task is to provide advice on
questions of local administrative reforms to the Chinese decision-makers.5The
program was initiated in 2000 and has taken place bi-annually since 2003.
Each time, the CCCPE receives 200–400 applications, and each application is
vetted for its eligibility to take part in the competition. CCCPE members
then make eld trips to those locations that are short-listed for the award, and
an expert committee consisting of roughly 12–15 distinguished Chinese scholars
selects the ten nal winners. By 2013, the program has received more than 2,000
5See http://www.chinainnovations.org (accessed on 21 January 2016).
6
applications, roughly half of which meeting the required quality criteria.6Given
the proximity of this organization to the apex of China’s political regime, the
Government Innovation Award is both well-known and considered prestigious
among local government ocials all over China. The incentives to apply for the
award are therefore high. The average application is 10 pages long, providing
basic information on the reform unit and lengthy answers to six open questions;
for example on the contents of the policy innovation, the reason for its initiation,
and its stage of implementation.
We applied a machine learning algorithm to the data in order to classify clus-
ters of co-occurring words into “topics” (Grimmer & Stewart 2013; Lucas et al
2015; Roberts et al 2016). The result is a so-called “topic model” (Blei et al
2003), in our case, it is composed of 100 topics, thus adequately mirroring the
content of the corpus. It must of course be noted that, above all, this is a com-
putation of general trends. However, by successively focusing on single parts
of the corpus, an improved representation can be achieved, even regarding such
topics that are underrepresented otherwise. By combining personal judgment
and computer-based analysis, the degree of abstraction can continuously be in-
creased or decreased. The resultant topics can then be coded numerically and
be analyzed statistically. To enable testing for the eects of the factors outlined
in the theoretical literature summarized above, we coded each case according to
its geographic location and added statistical information on factors such as the
per capita income when a reform was initiated.
The possibility of selection bias is one of the problems related to this kind of
non-arbitrary data collection. The question whether all politicians who have
devised a policy innovation are equally likely to apply for the innovation prize
is especially important. Theoretically, it is possible that the probability of an
application depends on the political and social networks a politician is part
of, on the information resources he disposes of, on his personality or on his
personal understanding of a policy innovation. Our relatively high number of
cases should however guarantee that such idiosyncratic factors are normalized
and will therefore not inuence the results signicantly. Moreover, the fact
that most of the initiators do not gather information about the prize from
social networks, but from the media, allows for the possibility to treat the
sample as representative regarding the geography and the institutional context
the applicant is situated in. This assumption is further strengthened by the fact
that almost all regions of China are equally represented in the sample. The wide
range of topics in the submitted applications further indicates an undistorted
sample of such measures that are regarded as political innovations by local
government ocials.
We consider the last factor – the subjective dimension of local policy innova-
tions – to hold the greatest potential for selection bias; less however for the
6The runner-ups and nalists are documented on the award home page and in a bilingual
publication, but the circulation of the whole set of applications is restricted to the CCCPE
researchers and the national committee that selects the nalists.
7
present analysis than for cross-country comparisons. Our research objects are
policies that local government ocials consider to be innovations, and the un-
derstanding of what constitutes a policy innovation is certainly inuenced by
cultural and political factors. For the present study, this is an advantage, as
we come to know about the social and political problems that, according to
local government ocials, require a solution. Moreover, we learn about the
types of solutions they deem not only feasible for local implementation, but
also compatible with the preferences of the central government - local ocials
usually justify policy innovations by referring to macro-policies and speeches
by China’s highest leaders. Among others, our data allows for conclusions on
whether the content of these proposals varies with a location’s economic devel-
opment and its geographic situation. Even in view of the possible limitations
outlined above, our contribution is important, none the least because it exploits
the most comprehensive and detailed dataset on policy innovations currently
available.
The Dynamics of policy innovation in China
In the following sections, we will apply the aforementioned methods to investi-
gate the major enabling factors for local policy innovations in China. With our
sample consisting of 1,024 cases, we have already determined that local policy
innovations are not deviant cases. Yet neither do we know whether these in-
novations actually took place in response to the hypothesised changes made by
the central government to the incentive structure of local ocials, nor whether
they are common to all regions in China. Theoretically, the observed policy
innovations could all cluster in a few select localities, which would cast doubt
on whether the central government’s policies indeed have a systematic impact.
We also know very little about what kinds of policy innovations local govern-
ments actually devise, and how frequently certain measures are in various parts
of China.
Policy innovations in time and space
In a rst step, we test whether the time of origin and the geographical distri-
bution of more than 1,000 policy innovations conrm the hypothesis that the
central government’s alteration of incentive structures faced by local govern-
ment ocials has exerted a systematic impact. If this hold true, we expect
a signicant increase in the number of policy innovations after 2004, the year
the Decision was implemented. Moreover, we expect to nd policy innovations
not only in prosperous regions with a high educational level, which is where
we would mainly expect them if incentives from the central government were
absent. We should rather be able to observe an even geographic distribution.
8
Figure 1: Policy innovations per year
0
25
50
75
100
1985
1990
1995
2000
2005
2010
Year
Cases
Regarding the initiation time of policy innovations, our expectations are not
fully met. Although the expected rapid increase in the number of local policy
innovations can indeed be observed in the years from 2005 to 2008, policy inno-
vations however existed prior to 2004. As gure 1 shows, measures have already
been devised and implemented proactively at a few locations in the 1980s. The
number of such locations is however low, corresponding to the fact that under
Deng Xiaoping, local governments had already been encouraged to act on their
own initiative, without, however, being punished if they failed to do so. The
fairly strong increase in local policy innovations from the end of the 1990s on
is however surprising – between 1997 and 1998, the number of local policy in-
novations almost increased threefold. A possible explanation is that those are
primarily economic measures implemented in response to the tax reform of 1996,
which required local governments to transfer irregular revenue into the regular
budget. The consequence of this reform was that they had to share parts of
their revenue with the central government and had to compensate their loss of
revenue by other means. This hypothesis will be tested in the fourth section of
this chapter.
The second question requiring clarication is concerned with the geographic
distribution of policy innovations. We have assumed policy innovations to be
spread all over China, but innovation density to be higher in the provinces
of the prosperous east coast than in the rest of China: they hold both the
necessary capital and the human resources needed to devise and to implement
policy innovations. It is furthermore probable that some cities have brought
forth more than one innovation, as they were able to refer to earlier experiences
for later innovations.
Figure 2 visualizes the geographic distribution of the policy innovations devised
in each municipal area. For reasons of anonymization and improved visibility,
9
Figure 2: Number of policy innovations per city
the gure does not distinguish between innovations brought forth on the com-
munal, the district, or the town level. Not all of the innovations visualized here
have therefore been devised by municipal governments.
The high geographical spread of policy is remarkable – in most of the cities, at
least one innovation has been devised. The provinces Tibet, Xinjiang, Gansu,
Qinghai, and Inner Mongolia are distinctive exceptions: they are marked by
their low economic level of development, a strong scal intervention of the cen-
tral regime, and partly by ethnic tensions. It should also be noted that inno-
vation clusters are by no means especially found at the east coast. A partial
exception is the Zhejiang province, where the majority of the cities has brought
forth policy innovations in the two-digit range. The city of Hangzhou leads
this ranking with 54 policy innovations. With 57 innovations, Peking is the
only city presenting a higher number. In western China, the cities of Kunming,
Chongqing, and Chengdu are especially innovative.
Types of policy innovation
In the following section, the analysis will subsequently be rened: not only the
fact that a policy innovation was produced, but also its content will now be
considered. As we have explained in the section on the methods employed for
this analysis, we have chosen an inductive approach to classify the documents.
10
By means of a machine learning algorithm, we have rst classied the more than
1,000 documents to such an extent that we can subdivide them into specic
measures. We extracted 100 topics, which, after excluding clusters of words
that do not signal any content and merging very similar topics, are condensed
to 55 topics. As gure 3 shows, these topics provide a very accurate insight into
the subject matter of the policy innovations implemented across China.
Figure 3: Policy innovation topics
0
20
40
60
vill_demo
soc_sec
pub_eval
admin_ref
particip
pubsec
soc_serv
adm_aprvl
land
vill_constr
mgrt_rgts
agri_prod
envrnmnt
construct
culture
prty_cnstrct
state_so
digitaliztn
ec_dev
soc_justice
soc_supprt
NPC
medicine
party_elct
pub_serv
youth
credit
hmownrs
mediate
open_gov
wmn_rght
educate
pub_opin
soc_partic
Intrnt_prtcp
brth_cntrl
cnflct_res
fiscal_ref
petition
philantrop
leaders
trnsprncy
urbn_comm
gov_effect
home_edu
vill_con
labor_rghts
desaster
hum_res
media_svs
admn_fines
mediate
anticorrp
pov_allev
pub_super
cooprtve
NA
issues
cases
Three topics dominate the sample: innovations regarding improvements to ru-
ral self-government, social welfare, and the evaluation of local administrative
performance by the general public. Ranked fourth and fth are improvements
to public administration (except for those to establish transparency and e-
ciency through public evaluation, see above) and the improvement of public
participation in local politics. As rural self-government is represented as a spe-
cic type, this category mainly refers to public participation at the municipal
neighborhood level. Again, our expectations are only partly met.
It should rst be observed that our sample displays the anticipated large num-
ber of measures devoted to improving social welfare and public administration,
implying that local governments are indeed eager to improve the communication
channels between state and society. The high prevalence of policy innovations
devised to improve rural self-government is however surprising. Equally sur-
prising is the prominence of measures in the category of public performance
evaluation of local administrations. The again unexpected and relatively high
number of measures to enhance public security is interesting. Such actions are
likely to constitute the counterpart to the proactive prevention of social unrest
by means of policy innovations in the eld of social policy. They are intended
11
to increase the ability of local governments to defuse threats to public security
that escape prevention. The 55 measures easily fall into nine superordinate
categories (gure 4):
Figure 4: Policy innovation typology
0
50
100
150
200
250
social_policy
particip
social_conflict
admin_reform
econ_policy
accountab
pol_reform
public_infrstr
environment
NA
issues
cases
Figure 4 once more illustrates that the majority of policy innovations in our
sample belong to the category of social welfare. The improvement of public
participation in local politics required to initially identify defects gure very
prominently in the sample as well.
These results support the hypothesis that the pressure the central government
exerts on the local regimes, for them to solve social problems on their own, actu-
ally has a systemic impact on local policy-making. Our expectations regarding
the prevalence of measures to decrease social conicts are met as well. By means
of these measures, particularly problems like the violation of migrant workers’
rights, of illegal land seizure by village and township administrations, and the
preservation of public security are supposed to be improved.
Administrative reforms are ranked fourth and are followed by measures intended
to improve the economic performance of a locality. The number of measures to
increase the transparency of public administration is equally high. Most of the
policy innovations summarized in this category provide for public participation
in the surveillance of administrative action.
Apparently, the majority of the policy innovations examined here is intended to
appease the population. It can thus be noted that local government ocials are
aware of the paramount necessity to be more ecient in solving social problems.
The formulation of measures for this purpose – whether to solve already existing
12
conicts, or to prevent conicts by providing social services more eciently –
suggests that the incentives implemented by the central government show the
desired eect.
Geographical inuences
In a third step, we want to determine whether the initiation of certain types
of policy innovation follows geographic patterns. As we have elaborated in the
theoretical section of this article, we expect two countervailing pressures to be
at work. On the one hand, the mandatory character of the policy innovation
imperative should provide for a wide geographical spread of policy innovation,
an expectation that has already been armed above. On the other hand, some
measures, in particular social welfare measures, tend to be more cost-intensive
than, say, administrative reforms, leading to our expectation of certain innova-
tion types appearing more frequently where the level of economic development
is high. We have indeed found some clusters of policy innovation in wealthy
regions, but clusters in less wealthy regions also exist. It remains to be tested
if these clusters dier, for example with social welfare measures clustering in
Beijing and Zhejiang and administrative measures in Yunnan and Sichuan.
Second, geographic factors apart from the economic development of a certain
location are expected to have an inuence as well: for instance, if developments
in China conform to the predictions of modernization theory, we should nd
measures aimed at improving participation opportunities especially in the more
developed areas. Third, as local governments compete against each other, towns
should respond to innovations initiated by their neighbors by bringing forth their
own policy innovations.
To determine whether this applies, we have visualized the distribution of the
aforementioned nine innovation categories on a map of China (gure 5). For
locations with more than one policy innovation, we have visualised the respective
rst relevant measure. Again, the distinct heterogeneity and absence of regional
patterns strikes at rst glance. Unexpectedly, the distribution of the individual
reform types appears to be quite arbitrary – in all parts of China, we nd each
type of innovation, with little regional accumulation of certain measures. There
is as little proof for our assumption of economic development and geography
favoring certain measures, as for the hypothesis that “contagion eects” emanate
from reforms initiated at a specic location. We have tested this by alternately
visualizing all 55 measures on a map. In only ve cases, the same measure has
been chosen in two neighboring villages –measures of the same type rather tend
to spread throughout China, with the partial exception of reforms to improve
the economic situation in rural areas: here, we we have found in Shandong
province an unusual conglomeration of four clusters of one measure implemented
in neighbouring localities.
The near total absence of geographic correlation is striking. Apparently, the
superordinate levels do not demand a policy innovation determined promising
13
Figure 5: Geographical distribution of policy innovation types
at one location to be implemented comprehensively in all other locations within
their jurisdiction. This either shows that the dynamics of policy innovation and
policy experimentation, which is driven by emulation, are entirely dierent, or
that none of the policy innovations was considered worthy of emulation. As
the policy innovations cited in the case studies above have been evaluated as
successful examples, the latter is unlikely. The more convincing explanation for
this phenomenon suggests that the appeal to devise new policy innovations is so
high that ocials refrain from emulating the measures of neighboring locations.
Timing and level of development
The analysis so far has delivered important insights regarding the impact of
various factors on the emergence of policy innovations in China. It has also
illustrated the wide range of measures covered by policy innovations. Further-
more, we found that policy innovations in China appeared a lot earlier and less
clustered than expected by us. Individual exceptions notwithstanding, we could
neither nd regional clusters of policy innovations in general, nor of specic
innovation types. Most of our expectations were conrmed only partly, if at all.
However, the analytical methods applied so far only allow for limited insights
into the impact of the dierent variables on the emergence of policy innovations.
Partly, this is due to the fact that we have primarily examined the status quo
14
and devoted little attention to sequencing. For example, the possibility that the
cities at the prosperous East coast have pioneered the development of policy
innovations, with all other cities only following suit after 2004 has not been
explored yet. This nding would go far towards explaining the presence of policy
innovations before 2004. Moreover, not all of the wealthy cities are located at the
east coast – it is quite possible that a higher income also raises the probability
of policy innovations in other regions. To examine the inuence of the timing of
a policy innovation and of the level of economic development on the emergence
of policy innovations, a dierent method of analysis is needed.
To test the impact of timing and economic development on the type of a policy
innovation, we will now perform a multinominal logistic regression. The depen-
dent variable is the type of innovation. We use the nine-part typology, because
a dependent variable consisting of 55 factors does not allow for robust results
in view of the relatively small per factor size of the sample. Two independent
variables have been selected: the year when the measure was initiated and the
natural logarithm of the gross regional product per capita in that year. To
exemplify the results of the model calculation to those of our readers without
prior statistical knowledge, we have converted the correlation coecients into
probabilities and we have visualized them accordingly.
Figure 6: Time, economic prosperity and policy innovation type
Year effect plot
Year
reftype2 (probability)
0.05
0.15
0.25
1990 1995 2000 2005 2010
: reftype2 accountab
0.05
0.15
0.25
: reftype2 admin_reform
0.05
0.15
0.25
: reftype2 econ_policy
0.05
0.15
0.25
: reftype2 environment
0.05
0.15
0.25
: reftype2 particip
0.05
0.15
0.25
: reftype2 pol_reform
0.05
0.15
0.25
: reftype2 public_infrstr
0.05
0.15
0.25
: reftype2 social_conflict
0.05
0.15
0.25
: reftype2 social_policy
LogGDPCap effect plot
LogGDPCap
reftype2 (probability)
0.05
0.15
0.25
−1 0 1 2 3 4
: reftype2 accountab
0.05
0.15
0.25
: reftype2 admin_reform
0.05
0.15
0.25
: reftype2 econ_policy
0.05
0.15
0.25
: reftype2 environment
0.05
0.15
0.25
: reftype2 particip
0.05
0.15
0.25
: reftype2 pol_reform
0.05
0.15
0.25
: reftype2 public_infrstr
0.05
0.15
0.25
: reftype2 social_conflict
0.05
0.15
0.25
: reftype2 social_policy
The left part of the gure displays the probability that a certain policy innova-
tion is implemented at a specic point in time; the right part the probability
15
of its initiation at a certain level of gross regional product per capita. To pre-
vent extremely low or extremely high income values from distorting the result
while at the same time taking into account the exponential increase inherent in
income variables, we follow econometric conventions and use the logarithm of
the GDP per capita instead of the absolute values.
The probability of a certain innovation’s implementation dependent on the two
factors time and income is indicated on the y-axis. This value should be self-
explanatory. It shows that innovations encompassing social welfare measures,
with around one fourth of all cases, appear most frequently in our sample. If our
sample is representative, then there is, on average, a 25 percent probability for
them to emerge in any location in the time span covered by our sample. This
value represents an average that neither accounts for the eect of the time of
initiation, nor for a locality’s GDP per capita. The degree of this average value’s
representativeness for the entire sample also remains unclear. An average value
is all the more signicant the less the sample’s individual values dier from it.
Figure 6 thus signies an improvement compared to these imprecise values:
it displays the probability of a certain type of measure’s initiation not as an
average, but depending on the two factors we are interested in. Moreover, the
grey-shaded line marks the 95 percent condence interval for the predicted value.
As expected, the results indicate that more prosperous locations were indeed
pioneers in terms of generating policy innovations in the social welfare domain.
The probability of an area initiating such a measure increases for higher levels
of GDP per capita, but decreases over time. It is thus plausible that policy
innovations initiated before 2004 were mainly of the social welfare kind and
initiated by prosperous cities. The aliation with an east-coast province is not
mandatory – we added a variable to the equation that measures this aliation,
and the resulting coecient even turns out to be negative – indicating that non-
east coast localities are more likely to engage in policy innovation than their
east coast counterparts.
An interesting nding is that while policy innovations in the social welfare do-
main decrease over time, measures to reduce social conicts increase to roughly
the same extent. This conrms our expectation that the central government’s
explicit fear of social unrest since 2004 is causing the generation of measures to
counter or reduce such conicts. Hence, the incentives of the central government
thus seem to work.
The majority of the other coecients is either very low, corresponds to the
predicted outcome, or remains unaected by either the initiation time or the
level of development. For example, policy innovations pertaining to the pub-
lic infrastructure are a rather recent phenomenon and more likely to appear
in less developed regions. Equally unsurprisingly, policy innovations targeted
at strengthening the economic performance are more prevalent in rather poor
regions. As predicted, innovations related to enhancing the eectiveness of local
public administration increase over the years. Surprisingly, measures aimed at
16
strengthening the accountability of local public administrations are not a recent
phenomenon, but have emerged rather consistently and independent of the level
of development. The same is true for innovations designed to enhance public
participation in local politics.
Other surprising ndings reach beyond the theoretical expectations in this paper
and deserve to be mentioned, yet need to be investigated in subsequent studies.
For instance, what surprises us is that innovations designed to improve local po-
litical structures and processes show a rising trend over time, but are less likely
for more prosperous localities. This contradicts modernization theory’s assump-
tion that political reforms are the result of economic development. The reverse,
but no less surprising eect can be observed for policy innovations related to en-
vironmental preservation. Although public dissatisfaction with China’s severe
environmental pollution has intensied, such measures tend to be pursued less
frequently in recent years. Surprisingly, the level of economic development does
not have any statistical eect.
Conclusion
Political scientists and the Chinese Communist Party seldom agree. Never-
theless, there existed a broad consensus among the two for many years, as they
both agreed that China’s local governments were the bottleneck of the country’s
social and political development. Even persistent critics of the regime shared
the party leaders’ assessment that corruption, maladministration, and the in-
competence of bureaucrats and politicians on the city, district, and town level
jeopardized the social stability of China. Although the guideline to “Strengthen
the Governing Capacity of the Communist Party of China”, published in 2004,
has attracted some scholarly attention, the implications of this guideline for
local leaders have remained under-researched.
A considerable number of publications question the prevalent paradigm of local
ocials being incapable on the basis of single-case studies, but do not provide
any information on whether the locations they investigated are deviant cases;
whether the paradigm itself is based on a distorted perception of the Chinese
reality; or whether the reforms, experiments, and policy innovations investigated
represent a change from the status quo, possibly induced by processes of learning
and adaptation.
The present analysis analyzed more than 1,000 local reform documents to an-
swer these questions. It yields some striking results, but does not allow for clear
conclusions: according to our sample, the number of local policy innovations
has indeed increased signicantly since 2004. However, we have also shown that
local governments have already been devising policy innovations since around
1998. This unexpected nding deserves further research. Moreover, the anal-
ysis has oered an insight into the thematic variety of policy innovations in
17
China and has shown that neither the level of economic development, nor ge-
ographic contagion eects can explain why at least one policy innovation has
been implemented in almost all Chinese cities.
Even if this diagnosis appears to be unsatisfying, the present article neverthe-
less builds a bridge between prominent and diametrically opposed positions;
namely of such research that considers local governments the weak spot of the
authoritarian one-party regime, and newer discourse claiming that local policy
innovation is a major source of its legitimacy. The nding that local policy
innovations have increased over of the years possibly reconciles these seemingly
antagonistic positions. We are somewhat cautious in voicing this assertion, be-
cause our ndings suggest that a process of local political renewal has indeed
been set in motion, but oer no clues regarding the sustainability of locally
initiated policies. The question where exactly the impetus for the increase in
local policy innovations in the late 1990s stems from also needs to be answered.
Further questions that should concern us are: why could so many hypotheses
deduced from the existing literature not be conrmed? Why are so many of the
outcomes examined in this study characterized by a near-complete absence of
regularities? One possible, but unlikely explanation for this diagnosis may be
that the majority of the measures investigated only exist on paper. A dierent
explanation appears to be more plausible: if structural and institutional factors
do not have a signicant inuence on the generation of local policy innovations,
then the latter might be explained by the personality and motivation of local
leaders. Apart from testing the results presented in this article, future studies
should also examine the impact of actor-related factors on the generation of
policy innovations, preferably by means of comparison of a large number of
cases. Herein lies one of the central challenges for governance research on China
in the next several years.
Bibliography
Blei, D.M., Ng, A.Y. & Jordan, M.I., 2003, Latent dirichlet allocation, the
Journal of machine Learning research, 3, pp. 993-1022.
Cao, C., 2000, Huanghe bian de Zhongguo (China along the Yellow River),
Shanghai wenyi chubanshe, Shanghai.
Chen, G. & Wu, T., 2004, Zhongguo nongmin diaocha (Investigation into the
Chinese peasantry), Renmin wenxue chubanshe, Beijing.
Chen, G. & Wu, W., 2007, Will the boat sink the water?: the life of China’s
peasants, PublicAairs, New York.
Chen, X. & Yang, X., 2009, Difang zhengfu gonggong guanli chuangxin: jingyan
yu qushi (Innovations in local government public management: experience and
tendencies), Jilin University Press, Changchun.
18
Florini, A., Lai, H. & Tan, Y., 2012, China Experiments: From Local Innova-
tions to National Reform, Brookings Institution Press, Washington.
Foster, K. 2004, Chinese public policy innovation and the diusion of innova-
tions: an initial exploration, in M Holzer, M Zhang & K Dong (eds), Frontiers
of Public Administration, The United Nations Public Administration Network,
New York, pp. 195-210.
Göbel, C., 2011, Uneven policy implementation in rural China, The China Jour-
nal(65), pp. 53-76.
Göbel, C. & Heberer, T. 2016, The Policy Innovation Imperative: Changing
Techniques for Governing China’s Local Governors, in V Shue & P Thornton
(eds), To Govern China: Evolving Practices of Power, submitted to Cambridge
University Press, .
Göbel, C. & Ong, L.H., 2012, Social unrest in China, Long Brieng, Europe
China Research and Academic Network (ECRAN).
Grimmer, J. & Stewart, B.M., 2013, Text as data: The promise and pitfalls of
automatic content analysis methods for political texts, Political Analysis(21),
pp. 1-31.
Heilmann, S., 2008a, From local experiments to national policy: The origins of
China’s distinctive policy process, The China Journal(59), pp. 1-30.
Heilmann, S., 2008b, Policy experimentation in China�s economic rise, Studies
in Comparative International Development, 43(1), pp. 1-26.
Heilmann, S., 2009, Maximum Tinkering under Uncertainty Unorthodox
Lessons from China, Modern China, 35(4), pp. 450-62.
Heilmann, S. & Perry, E.J. 2011, Embracing Uncertainty: Guerrilla policy style
and adaptive governance in China, in Mao’s Invisible Hand: The political foun-
dations of adaptive governance in China, Harvard University Press, Cambridge,
MA, pp. 1-29.
Hillman, B., 2010, Factions and spoils: examining political behavior within the
local state in China, The China Journal(64), pp. 1-18.
Kennedy, J.J., 2009, Maintaining Popular Support for the Chinese Communist
Party: The Inuence of Education and the State-Controlled Media, Political
Studies, 57(3), pp. 517-36.
Li, L., 2002, Wo xiang zongli shuo shihua (I told the truth to the premier),
Beijing: Guangming ribao chubanshe.
Li, L., 2004, Political trust in rural China, Modern China, 30(2), pp. 228-58.
Li, L., 2008, Political trust and petitioning in the Chinese countryside, Compar-
ative Politics, 40(2), pp. 209-26.
19
Lucas, C., Nielsen, RIchard A., Roberts, Margaret E., Stewart, Brandon M. &
Tingley, Dustin, 2015, Computer assisted text analysis for comparative politics,
Political Analysis, 23(2), pp. 254-77.
Roberts, M.E., Stewart, B.M. & Tingley, D. 2016, Navigating the local modes of
big data: The case of topic models, in Computational Social Science: Discovery
and Prediction, Cambridge University Press, Cambridge, pp. Chapter 2.
Shambaugh, D.L., 2008, China’s communist party: atrophy and adaptation,
Univ of California Press, Berkeley et. al..
Smith, G., 2009, Political Machinations in a Rural County, The China Journal,
62(July 2009), pp. 29-59.
Smith, G., 2010, The Hollow State: Rural Governance in China, The China
Quarterly, 203, pp. 601-18.
So, A.Y., 2013, Class and Class Conict in Post-socialist China, World Scientic,
Singapore.
Teets, J.C. & Hurst, W., 2014, Local Governance Innovation in China, Rout-
ledge, London/New York.
Zhong, Y., 2015, Local government and politics in China: Challenges from
below, Routledge, Abingdon.
Zhu, X., 2014, Mandate versus championship: Vertical government intervention
and diusion of innovation in public services in authoritarian China, Public
Management Review, 16(1), pp. 117-39.
Zhu, Y., 2012, Policy entrepreneur, civic engagement and local policy innovation
in China: Housing monetarisation reform in Guizhou Province, The Australian
Journal of Public Administration, 71(2), pp. 191-200.
20