The misruling elites: the state, local elites, and the social geography of the Chinese Revolution

Abstract

The existing scholarship has developed six main explanations to account for the success of the Chinese Revolution, which has been anomalous for major paradigms derived from cross-national comparisons. Methodologically, we use a social geographical approach to test these existing explanations systematically by constructing and analyzing a unique dataset of Communist growth in 93 counties in the three most contested provinces during its most pivotal period of ascendence. Theoretically, we advance and test an alternative perspective, based on the groundwork of Tocqueville and Fei Xiaotong, that integrates the state-centric theory, elite theory, and cultural analysis. Our perspective emphasizes the interplay between state centralization and local elite structure, which leads to intensified state extraction and local elite fracturing, thus creating favorable conditions for revolution. The quantitative analysis strongly supports the importance of the Japanese invasion but provides limited support for many other conventional explanations. The analysis largely confirms the Tocqueville-Fei perspective on state centralization, elite fracturing, cultural change, and revolution. The findings are buttressed by a detailed case study of Lianshui County. The study unveils a common structural challenge that a modernizing state faces in an agrarian status society, to recreate its political legitimacy while disrupting local elite structure. It also sheds historical light on the evolution of state-society relationship through the Chinese Revolution.

The Chinese Revolution is one of the most consequential revolutions in modern history. Why it succeeded, whereas similar Communist attempts failed in many other countries, remains a subject of enduring debate. In particular, the Chinese Communist Party’s (CCP) dramatic turnabout during China’s war of resistance against Japan (1937–1945) from being on the brink of total collapse to being poised to take state power is a puzzle that generations of scholars have tackled. Existing explanations emphasize a range of factors, including intensified rural class conflict (Chen, 1933, 1936; Wolf, 1969), the weakening of the Kuomintang (KMT) state’s military power by the Japanese invasion (Mann, 2012; Feng & Goodman, 2000), “peasant nationalism” provoked by foreign invasion and tapped into by the CCP (Johnson, 1962), the CCP’s socioeconomic reforms that won over peasant support (Selden, 1971), the CCP’s exceptional mobilizational capacity (Moore, 1966; Skocpol, 1979, Chen, 1986), and the “social banditry” tradition that the CCP was able to leverage (Hobsbawm, 1974; Skocpol, 1979; Van Slyke 1986).

This article, on the one hand, systematically assesses these existing explanations with a methodological innovation of studying changes in CCP power over time within counties, the basic unit of governance in Imperial China and the site of the most consequential changes in the state-society relationship in the modern era. On the other hand, it develops and assesses an alternative perspective that challenges the assumption of local elite cohesion in the existing explanations and examines the interplay between the state and local elite structure.

Methodologically, we extend the social geographical approach that John Markoff’s (1985, 1986) analysis of peasant revolt in the French Revolution pioneered (more recent works in this vein include Mazur, 2019 and Barrie, 2023). We aim to study revolutions akin to electoral geography in democracies, positing that variations in local social conditions can account for differences in revolutionary outcomes. We collect and combine data from multiple sources on the historical record of 93 counties in Henan, Anhui, and Jiangsu provinces in Central China in 1937–1945, when Communist guerrilla power experienced the most dramatic growth. During the period, these three provinces were also the most heavily contested between the Communists, the Kuomintang Government (KMT), and the Japanese military forces.

Theoretically, we expound and test an alternative perspective grounded in the analysis of the relationship between state centralization, local elite fracturing, and revolution that Alexis de Tocqueville and Fei Xiaotong advanced regarding revolutions in France and China, respectively. This Tocqueville-Fei perspective challenges the assumption of local elites as a coherent ruling class in the existing literature and puts the “groupness” of local elites in a revolutionary situation under critical scrutiny. It integrates the state-centric theory of revolution (Foran, 2005; Goldstone, 1991; Goodwin, 2001; Skocpol, 1979), the focus on elite fracturing and political change (Bearman, 1993; Gould, 1996; Lachmann, 2003; McAdam, 1996; Mizruchi, 2013; Zhang, 2021), and the culturalist approach to revolution (Arjomand, 1989; Keddie, 1995; Sewell, 1985, 1996). It argues that state centralization unsettled the checks and balances between the central state and traditional local elites, and replaced the dominant but protective traditional elites with new state agents and brokers who were often predatory and abusive of local peasantry. The consequence is that state centralization increased both state extraction of local resources and local oppression, thus increasing the susceptibility to revolutionary takeover. The strength of traditional elite legacy, on the other hand, could reduce state extraction, which would in turn decrease the receptiveness of revolution, but could also increase elite defection to revolution.

The results of our analysis indicate that Japanese invasion was a strong predictor of Communist takeover in the following year (Mann, 2012). They also suggest Communist capacity in carrying out socioeconomic reforms (Selden, 1971) and civilian mobilization (Moore, 1966; Skopcol 1979, Chen, 1986) were significant factors. However, they show little support for three conventional accounts, ranging from the orthodox Marxist theory of intensified rural class conflict, to arguments about Communist leadership of “peasant nationalism” (Johnson, 1962), to the “social banditry” interpretation (Hobsbawm, 1974; Skocpol, 1979; Van Slyke 1986).

In contrast, our results largely substantiate the Tocqueville-Fei perspective on the relationship between state centralization, local elite fracturing, and revolution. In the face of Japanese invasion, counties with deeper state penetration were more likely to experience a Communist takeover, so were those with higher levels of state extraction of local resources. However, while both state centralization and excessive resource extraction were strong predictors of CCP takeover, state centralization did not increase CCP takeover through excessive resource extraction. This suggests that much of the effect of state centralization on CCP takeover was mediated by local elite fracturing, specifically through disrupting local elite structure with newly empowered state agents. On the other hand, counties with a stronger gentry culture were more likely to experience a Communist takeover. However, gentry culture legacy did not decrease CCP takeover through directly moderating state extraction of local resources. This indicates that much of the effect of gentry culture legacy on CCP takeover was mediated by local elite fracturing, particularly through elite defection from the KMT state to the CCP revolution.

A limitation of our multivariable analysis is the absence of network data to measure elite fracturing. Accordingly, we supplement the statistical analysis with a case study of Linshui County to illuminate how the two contrasting mechanisms of elite fracturing, entailed by state centralization on one hand and local gentry legacy on the other, interacted and played out on the ground. Taken together, the multivariable analysis and case study raise important theoretical implications about social formation of elites in the transitions to modernity and the role of culture in elite conflict and revolution, and shed new light on the long-term evolution of state-society relationship during and after revolution.

Explaining the anomalous case of the Chinese Communist Revolution

The success of the Chinese Communist Revolution, especially its spectacular development during China’s Resistance War against Japan (1937–1945), has provoked generations of explanations, in both comparative historical research and China studies. By 1934, the Communist movement in China had almost been annihilated by the KMT’s crackdown. It was only during the war that the Communist forces experienced exponential growth. In 1937, the CCP-controlled military forces numbered fewer than 40,000 and its party membership was a little over 40,000. By April 1945, its military amounted to 910,000 and its party membership to 1,210,000. The population under its rule, concentrated in North and Central China (including the three provinces we examine), reached 100 million (Van Slyke 1986:709). By then, the Communists were already strong enough to contend for state power. After the American effort to broker peace between the KMT and the CCP failed in 1946, the CCP took only three years to defeat the KMT nationally.

How to explain this success? So far, scholars have found the Chinese Revolution a significant anomaly for the major theoretical paradigms derived from cross-national comparisons. The revolution was not preceded by rapid capitalization of rural landownership and urbanization, as modernization theory posited (Huntington, 1968; also Tilly, 1964, Moore, 1966). Agricultural commercialization began in China in the twelfth century and was already a fait accompli by the seventeenth century, especially in Central and Eastern China (Hung, 2008; Pomeranz, 2000). Chinese peasants were not caught up in dependent development, as the neo-Marxist paradigm assumed (Wolf, 1969; Paige, 1975; also, Foran, 2005).

Both Moore (1966) and Skocpol (1979) admitted that Chinese peasants, unlike the French or Russians, lacked the long tradition of community solidarity and autonomy, an important condition in their groundbreaking comparative works. More recent state-centric theorists (Foran, 2005; Goodwin, 2001), who consider exclusionary and personalist regimes, weak state capacities, and economic downturn as important conditions for revolutions, also have difficulty coming to terms with the facts that the impact of economic downturn on Chinese peasants was quite immaterial (Rawski, 1989; Mann, 2012: 419–21) and the revolution succeeded precisely during the KMT rule, when the state was less exclusionary, stronger, and more bureaucratically rational than the preceding warlord government.

The existing scholarship has developed six explanations to account for this Chinese anomaly. First, drawing on his IEMP (ideological, economic, military, and political) model of the four sources of social powers, Mann (2012) contended that existing theoretical paradigms in the sociology of revolutions focused too much on ideological, economic, and political powers and overlooked military power. He pointed out the connection between the Sino-Japanese war and the CCP’s growth and argued that the Japanese invasion considerably eroded the KMT state’s military capacity to repress revolution and provided fertile ground for the growth of Communist guerrilla forces.

The second explanation, known as the thesis of “peasant nationalism” advanced by Johnson (1962), draws a different connection between the war and the CCP’s growth. Johnson argued that when the Japanese invaded peasant communities, the CCP provided peasants with leadership to fight the invaders that the KMT lacked and thus won over peasant support (an argument that struck a chord amidst the increasing U.S. involvement in Vietnam in the 1960s).

Third, Selden (1971) maintained that what won over the peasants was not nationalism but tangible benefits that the CCP offered. Based on his study of the Communist base areas in the northwest provinces of Shaanxi, Gansu, and Ningxia, he argued that peasants were less nationalistic and more sensitive to their own material interests and contended that it was the Communists’ socioeconomic reforms such as land, tax, and rent policies that attracted peasant support (also, Kataoka, 1974; Wou, 1994).

The fourth model, developed by Sino-Marxists (Chen, 1933, 1936; China Institute of Pacific Relations, 1938), posited that Chinese peasants were increasingly proletarianized, thus becoming attracted to revolts. The crucial indicator, they argued, was land tenancy. That is, the more peasants were tenants rather than self-tillers, the more intense their revolutionary grievances would be and the more likely they would embrace the Communists who advocated for land reform, because the Chinese tenants, in addition to being agricultural wage laborers, were also personally dependent on their landlords and thus suffered the worst exploitation and oppression. This Sino-Marxist emphasis on intensified rural class conflict was also echoed in the neo-Marxist and neo-Weberian scholarships on peasant revolutions (Moore, 1966; Paige, 1975; Skocpol, 1979; Wolf, 1969).

Fifth, Moore (1966) and Skocpcol (1979) resolved the Chinese anomaly in their cross-national comparison by emphasizing the exceptional leadership and mobilization that the CCP provided, which presumably compensated for the lack of peasant solidarity and autonomy. This emphasis on the CCP’s exceptional mobilization was reinforced by Chen (1986), who drew on the CCP’s internal sources to document Communist activity at the local level in Anhui and Jiangsu provinces to highlight Communist mobilizational astuteness (also, Wou, 1994).

Finally, Skocpcol (1979) further built on Eric Hobsbawm’s (1974) notion of “social banditry,” by which she meant that Chinese Communists tapped into an endemic tradition of unruly peasants fleeing to the mountainous areas and accumulating revolutionary strength from the peripheries of rural China. Van Slyke (1986: 652) echoed this thesis by claiming that terrain mattered for the Communist success as they seemed to do better in mountainous and remote areas.

These six explanations in historical social sciences, China studies, and from within the Chinese state itself are fascinating and have occasioned much useful research and debate. One commonality these explanations share is that their main method is qualitative cross-national comparison or case studies. They have not been systematically tested. One goal of our study is to test these explanations quantitatively through examining the subnational variations of CCP power within the same counties over time.

The other shared assumption between these explanations is that local elites were a coherent ruling class-for-itself. This assumption, however, contradicts the emergent insight from contemporary elite theorists that elite conflict and fracturing is a necessary political opportunity for political change (Bearman, 1993; Burton & Higley, 1987; Gould, 1996; Lachmann, 2003; McAdam, 1996; Mizruchi, 2013; Zhang, 2021). Building on this insight, we develop a perspective, as the second goal of our study, that puts the changing “groupness” (Brubaker, 2004; Tilly, 1978: 62) of local elites under critical scrutiny. This perspective takes Goodwin (2001)’s state-constructionist approach one step further by unpacking the impact of state centralization on the social formation of local elites as well as the influence of local elite structure on revolutionary outcome.

The Tocqueville-Fei perspective on the state and local elite structure in revolution

We come to this perspective from the groundwork of Alexis de Tocqueville and Fei Xiaotong. These scholar-politicians hailed from very different contexts, yet both examined revolution in light of the tension between state centralization and local elite status culture, the fracturing between old local elite and new state agents, and the political opportunity these two conditions offer to revolutionaries.

In The Old Regime and the French Revolution, Tocqueville (1983) argued that, under the Old Regime, local autonomous institutions led by the landed nobilities had been gradually replaced by the absolutist state bureaucracy led by intendants and subdelegates, who were men “of common birth”. As the nobilities were deprived of their former powers while maintaining their privileges, they—except those in the Vendée (p.122)—now rescinded noblesse oblige, their moral responsibilities to protect peasants. While in theory, the King’s Council was to shoulder these responsibilities, the practices of office venality and tax farming often rendered the peasants in a state of neglect (p.40, 120–137). This then created one of the conditions for revolution.

Fei Xiaotong (1910–2005), a foundational figure of Chinese sociology and anthropology, developed a parallel analysis with regard to the Chinese Revolution, apparently without being aware of Tocqueville’s.¹ They shared a similar focus on the tension between state centralization and local elite status culture and the ensuing elite fracturing in local governance, despite the significant differences between these two cases in the specificities of the dynamics between state centralization, change in local elite structure, and revolutionary outcome. During World War II, Fei spent years studying the changes in local administration in order to understand the ongoing rise of Chinese Communism. His observations were later published by Chicago anthropologists, Robert and Margaret Redfield, when Fei was suffering difficulties in his academic and political career in China.

In his book China’s Gentry, Fei (1953) observed that the KMT’s state centralization had unsettled the traditional checks and balances between the imperial state and local elites—that is, the Confucian gentry. According to him, these two powers, as illustrated in Fig. 1, were traditionally interdependent while also maintaining checks and balances against each other, even though these checks and balances were, by modern democratic standards, limited. The gentry were typically imperial degree holders or retired officials, whose power in local communities hinged on the honor conferred by the state. On the other side, the reach of the imperial state extended only so far as the county magistrate, who, as an outsider appointed for a limited term, needed the cooperation of the local gentry to govern effectively (also, see Qu, 1962; Esherick & Rankin, 1990). Similar systems of the balance of power between the state and local elites also developed in other great agrarian empires and states, ranging from Western Europe (Bloch, 1961), the Ottoman Empire (Barkey, 2008) to Japan (Smith, 1959).

Fig. 1

Traditional state-local-society relationship

According to Fei, when the modernizing KMT state increasingly penetrated local society, it disrupted this intricate system of checks and balances as it attempted to collapse the different roles of gentry, community representatives, and county runners (yayi, grassroots-level staff), whose social statuses were worlds apart, into a clear hierarchy of state agents. Very much like their French counterparts, many Chinese gentry, whose social status was previously equal to the county magistrate, chose to stand down or exit in order to avoid demeaning themselves into the latter’s underlings. County runners and community representatives, whose power was previously limited by low social status and formal institutional constraints (e.g., restriction from civil service examinations for three generations), were able to aggrandize themselves on the back of the coercive power of the centralizing state. As local norms of reciprocity were replaced by exploitative brokerage predicated on the state machinery, local elites became increasingly fractured between old gentry leaders with residual informal prestige and new state agents with formal powers; state agents became much more predatory; and the tax burden became increasingly onerous, such that peasant grievance became a galvanizing force for the Communists to tap into. For Fei, the rapid rise of Chinese Communism during the Japanese invasion needs to be understood in reference to this long process of fracturing and predation.

Michael Mann’s (1986) IEMP model can help us understand this change. Traditionally, the Confucian gentry held all four sources of social power in local communities: ideological (honor and prestige), economic (big landownership), military (informal power to organize militia during dynastic decline and unrest), and political (informal leadership). Their possession of these powers locally was nevertheless checked by and interdependent with the imperial state, through civil service examinations (ideological), property system and tax and levy exemptions (economic), imperial garrisons (military), and civil bureaucracy down to county magistrates (political). As the modernizing state centralized and penetrated local communities through formal administrative expansion, the gentry retained much ideological power, whose status honor had inhibited them from taking formal positions subordinate to the county magistrate. The local military and political powers now shifted to the new state agents, who took advantage of the state’s coercive power to aggrandize themselves, including economically. Culture, specifically the traditional elite status culture and noblesse oblige and the deficiency of local legitimacy of the new state agents, thus played a differentiating role in the fracturing between the old and new local elites.

Admittedly, the French Revolution and the Chinese Revolution differed remarkably in local elites’ composition, structural change, and political realignment, and Fei’s observation speaks more closely than Tocqueville’s to the specific mechanisms of the Chinese Revolution. Nevertheless, what we mean by a shared Tocqueville-Fei perspective is their threefold focus on state action, local elite structure, and cultural change: namely, the impact of state centralization on the local elite structure and the role of culture in shaping this impact. While each of the three components has been emphasized respectively by the state-centric perspective, elite theory, and cultural analysis, it is the dynamic relationships between them that the Tocqueville-Fei perspective captures vividly and comprehensively.

State-centric theorists have long recognized that state centralization can increase the likelihood of revolution by increasing its extraction of local resources. As Tilly (1985)’s characterization of the state as extortion or protection rackets suggests, state centralization is accompanied by more extraction of societal resources, to a great extent to face international military competition. John Markoff (1985)’s pathbreaking multivariable study of the social geographic patterns of peasant revolts in the French Revolution (also, Markoff, 1986) confirmed this correlation between state centralization and peasant revolts. However, state-centric theorists often treat local elites as coherent ruling classes and relegate culture as epiphenomenal to the material basis of revolution (Skocpol, 1979).

Contemporary elite theorists are critical of economic and class-based determinism and insist on the independent role of elite structure, especially the political ramifications of elite unity, elite conflict, division, and fracturing, and “elite settlement” (Bearman, 1993; Burton & Higley, 1987; Gould, 1996; Lachmann, 2003; Mizruchi, 2013; Zhang, 2021). In this vein, elite fracturing and division is recognized as a structural precondition of revolutions and other political upheavals as it creates a structural opportunity for elite defection to insurgent movements or structural reforms (McAdam, 1996). Yet, as Laura Edles (1995) points out, contemporary elite theorists often treat elites as rational choice actors, without regard to their cultural contexts. Changes in elite structure, she argues, are culturally embedded and enabled processes.

Our synthesis of Tocqueville and Fei integrates this culturalist lens with the insights from the state-centric theory and elite theory and maintains that the interplay between state centralization and local elite structure is a culturally conditioned and mediated process. That is, culture exerts influence on local elite politics both as the initial condition of elite norms, or lack thereof, and as the meaningful process of elite fracturing, normative disintegration, and defection. This Tocqueville-Fei perspective offers a set of dynamic and empirically testable hypotheses, as we focus on the changes at the county level, the basic unit of China’s local administration and community.

First, agreeing with the state-centric theory, the Tocqueville-Fei perspective posits that,

• Hypothesis TF1: Counties with higher state centralization are more likely to exhibit increased state extraction of local resources, which in turn, increases the probability of revolutionary takeover.

Second, the Tocqueville-Fei perspective argues that excessive extraction of local resources is not the only causal pathway by which state centralization increases the probability of revolutionary takeover. The other causal pathway is through changing the local elite structure. Specifically, state centralization upset the traditional system of checks and balances, in which norms like noblesse oblige constrained Confucian gentry and status differentials in turn constrained their local non-gentry underlings like county runners and community representatives. When the modernizing state penetrated local communities by mobilizing local non-gentry agents who were previously constrained under the traditional system of checks and balances, it unleashed these new and predatory local elites, whose abuse of power aggravated peasant grievances. This mechanism has been underscored by Duara (1991), whose critically acclaimed study of six villages in North China was in part influenced by Fei’s China’s Gentry. Hence, the Tocqueville-Fei perspective contends that,

• Hypothesis TF2: Counties with higher state centralization are more likely to experience increased local elite fracturing and accentuated local oppression, which in turn, increases the probability of revolutionary takeover.

Third, the Tocqueville-Fei perspective suggests that the interplay between state centralization and local elite structure is shaped not only by the intensity of state centralization, but also by the initial condition of local elite structure, in particular the local endowment of Confucian gentry culture. A stronger legacy of Confucian gentry culture means a stronger presence of traditional elite noblesse oblige to protect local community and a stronger resistance to state extraction of local resources, thus mitigating the local discontent that would be favorable for revolution. Thus, the Tocqueville-Fei perspective posits,

• Hypothesis TF3: Counties with strong local gentry legacy are more likely to exhibit reduced state extraction of local resources, and hence decreased probability of revolutionary takeover.

Finally, a stronger legacy of local Confucian gentry culture, on the other hand, can also mean increased elite fracturing, because the remaining traditional gentry elites would have a stronger obligation to distance themselves from the new and predatory state agents. This increased elite fracturing and division then increases the chance of revolution, as the remaining local gentry may defect from a corrupt and abusive regime for an insurgent revolutionary movement. This hypothesis is the most contradictory to the CCP’s official narrative of a coherently oppressive landed class, a narrative that has also been reproduced in the Marxist and neo-Weberian scholarship (Skocpol, 1979). For example, Wolf (1969: 132) draws upon Chinese Communist ideologue Chen’s (1933: 18) notion of “quadrilateral beings” of “rent collectors, merchants, usurers, and officials” to describe the Chinese local elites as a coherent landed class. Yet, the Tocqueville-Fei hypothesis of gentry defection has been corroborated by Benton’s (1999) study of the Communist-led New Fourth Army in 1938–1941 in part of Central China. He documents how Communist guerrilla leaders curried favor with local gentry leaders by addressing them by their scholarly names and composing classical poems in correspondence with them to prove cultural affinity, which helped to sway local gentry from the KMT regime and form an important backbone of Communist growth, largely because they were already deeply disaffected with the KMT. To sum up,

• Hypothesis TF4: Counties with strong local gentry legacy are more likely to exhibit increased local elite fracturing and heightened gentry disaffection and defection, and hence increased probability of revolutionary takeover.

Figure 2 captures these hypothesized mechanisms.

Fig. 2

Tocqueville-Fei hypotheses: State centralization, local elite structure, and revolution

The Tocqueville-Fei hypotheses imply the following corollaries: (1) the greater state centralization is in a county, the more likely the county is to experience revolutionary takeover, (2) the greater the state extraction of local resources a county endures, the more likely the county is to experience revolutionary takeover, and (3) the more fractured the local elite is in a county, the more likely the county is to experience revolutionary takeover.

While the effect of local gentry culture legacy on revolutionary takeover goes into opposite direction in the two pathways, given the stronger emphasis that Tocqueville and Fei placed on the mechanism of local elite fracturing, we may also expect that (4) the stronger the local gentry culture legacy is in a county, the more likely the country is to experience revolutionary takeover.

Data and methods

In the long history of Imperial China, the county was the basic unit of government administration and, since the turn of the twentieth century, has been the prime site of changing state-local society relationships, as lower administrative units (township and village governance) were gradually created and institutionalized. Accordingly, we conduct the empirical analysis by county and year.

We focus on three provinces in Central China—Henan, Jiangsu, and Anhui—for three reasons. First, this area was the most contested during the war between the Japanese, the CCP, and KMT. By comparison, up in the north, the pattern of Japanese occupation of cities and CCP control of the countryside had already stabilized after 1938. Down in the south, the contestation was mainly between the Japanese and KMT, with minimal CCP guerrilla bases.² Second, they are the context of two major studies at the local level: Chen (1986) on Anhui and Jiangsu, and Wou (1994) on Henan. Finally, the terrain of Henan, Anhui, and Jiangsu are diverse, with flat plains in the basins of the Huai River and tributaries near the coast, and mountains in western Henan and southern Anhui. This topographical diversity enables us to evaluate Van Slyke’s (1986) topographical extension of Skocpol’s “social banditry” thesis.

We focus on the period of Japanese invasion, not only because it is the most consequential phase of Communist growth and is hence the focus of the existing literature, but also because the pattern of Communist growth in this period was indisputably local. Typically, the CCP would send a small team with local ties to a county to develop a guerilla base based on the manpower and resources they could find on the ground, with much policy guidance but minimal material assistance from the party center. By comparison, during the subsequent Civil War (1946–49), the CCP’s military was so strong that it often overrode local factors and invaded counties where it had no guerilla base at all. Because of the pattern of local-based growth in 1937–1945, our subnational comparison of counties can clearly detect the relation between local socio-structural conditions and the growth of Communist power.

To sum up, we aim to explain the growth of Communist control in 93 counties in these three provinces in Central China during the Japanese invasion from 1937 to 1945. To our knowledge, our study is the first multivariable longitudinal analysis of any revolution at the local level that accounts for unobserved differences among the localities.³

Data and variables

To capture within-county over-time variations across multiple counties, we take county by year as the unit of analysis and operationalize our concepts with a series of variables that vary by county and year, from 1937–1945.

The dependent variable is the presence of a CCP County Committee. We use County Committee as the indicator of CCP control because, according to principles documented in the CCP’s official organizational history, in a county, the party organization grew from cell groups and party branches through intermediate forms such as executive committee (zhiwei), action committee (xingwei), working committee (gongwei), special committee (tewei), and temporary committee (linwei), culminating in the formal establishment of the County Committee (xianwei), which took place when the Communists had already stabilized their control over a base area. And when such control was impaired, the County Committee was usually abolished (MOHCCP, Vol.3, 2000: p.257, 853; Li, 2010).

We draw information about presence or absence of CCP county committee in a given year from county gazetteers, a rich and reliable source that social scientists have increasingly used for quantitative historical analyses in recent years (Su, 2011; Walder & Lu, 2017; Javed, 2017; Sng, et al., 2018; Lu et al., 2020).⁴

We operationalize the six conventional explanatory factors in the following ways:

• (i) to measure the effect of Japanese invasion (Mann, 2012), we identify whether the Japanese had occupied the county seat at any time during the year, since this was the main form of Japanese invasion.

• (ii) to measure CCP leadership in tapping into “peasant nationalism” and fighting the Japanese invaders (Johnson, 1962), we record instances of CCP-led armed combat and resistance against the Japanese invaders and their Chinese collaborators and CCP military enlistment and conscription during the year.

• (iii) to operationalize CCP socioeconomic reforms (Selden, 1971; also Kataoka, 1974, Wou, 1994), we record implementation of rent reduction, interest reduction, and tax reform.

• (iv) to operationalize CCP mobilization (Chen, 1986; Moore, 1966; Skocpol, 1979), we record instances of CCP mobilization of civilians through the establishment of civil organizations, raising funds and organizing voluntary donation of grains.

The above four variables are all time-varying, whose data, like CCP control, are drawn from county gazetteers.

• (v) to test the “social banditry” thesis (Hobsbawm, 1974; Skocpol, 1979; Van Slyke 1986), we obtain information on terrain from China historical GIS datasets at Harvard University and construct the ruggedness of the terrain of each county (Fernández-Villaverde et al., 2023). Ruggedness is measured by the standard deviation of elevation. This variable is not time-varying.

• (vi) to test the effect of intensified rural class conflict (Chen, 1933, 1936; Wolf, 1969), we follow Sino-Marxist emphasis on tenancy rate. Theoretically, tenancy rate was time-varying. However, there was no survey of tenancy rate during the war time of 1937–1945. The most systematic survey data of tenancy rate came from a major investigative initiative by the KMT government in 1932–1933. The context is that, by the early 1930s, taxes borne by peasants reached such levels that the plight of peasants had become a national concern. The KMT state conducted these investigations in preparation for major reforms of local taxation and administration (Liu, 1935). Although these reforms did not materialize due to the disruption of the war, the investigations left the best available statistical data. We draw these data, including those on tenancy rate, from a 2,266-volume administrative record entitled Collection of Historical Materials on Republican China and its sequel (Minguo Shiliao Congkan/xubian; hereafter, CHMRC/sequel), compiled by a team in Beijing Normal University.⁵ Since the data on tenancy rate only came from 1933, tenancy rate is also not time-varying. We will discuss our modelling strategy for non-time-varying variables below.

In Table 1, the first six rows, summarize how we operationalize the six conventional explanations, data sources, and the arguments. The last four rows, on the other hand, outline our operationalization of the four Tocqueville-Fei hypotheses.

• (1) To represent state centralization, we extract data on KMT-organized local militia, and population from the CHMRC/sequel collection. Then, for each county, we compute the ratio of local militia to the rural population. After gaining power in 1927, the KMT state implemented the baojia reform to penetrate local communities and a central pillar of this reform was to organize peasants into KMT-controlled militia. The militia was intended to discipline the population, make peasants more responsive to state mobilization, and help prevent Communist influence, banditry, and other disruption. While in the scheme of baojia reform, most able-bodied men should have been recruited into the militia, it proved to be a daunting task due to the weak “infrastructural power” of the state (Mann, 1984). Despite years of effort to enforce militia recruitment, by 1933, an average of just 12.27 per thousand rural residents, with standard deviation of 29.29, had been recruited (see Table 2). In this sense, the ratio is a reliable indicator of state centralization.

• (2) To measure gentry culture legacy, we draw on data of “Imperial Scholars” (jinshi) in the Index to Stelae Inscriptions of Imperial Scholars in Ming and Qing Dynasties (mingqing jinshi timing beilu suoyin) (Zhu & Xie, 1998), to construct a variable to represent the strength of the gentry culture legacy. This is based on the number of persons from the county awarded the title of Imperial Scholar in the Imperial civil service examinations between 1864, the end of the Taiping Rebellion during which local elites in Central China were severely disrupted, and 1905, when the examinations were abolished and replaced by Western-style education. The measure of gentry legacy is the number of Imperial Scholars divided by the county population as reported in the CHMRC. The mean number is 19.21 per million population with a standard deviation of 33.65.

• (3) To represent excessive resource extraction over local community, we use the survey data on taxes in 1933 in the CHMRC/sequel collection, to compute the ratio of tax surcharges (tianshui fujia) over regular land taxes (tianfu zhengshui). Historians have shown that, since the Qing through to the period of study, regular land tax remained steadily low. Yet, following the fiscal crisis incurred during the disastrous First Sino-Japanese War (1894–1895), tax surcharges were imposed and gradually increased in order to pay for war debts and finance modernization of the military, infrastructural projects, and education (Zhou, 2000). The tax surcharges remained reasonable up until 1927, since the gentry-dominated local assemblies strove to defend local interests and oppose onerous tax extraction (Li, 2005; Rankin, 1986). However, after the KMT state implemented top-down centralization, taxes began to soar. Because regular land taxes remained modest and steady, whereas tax surcharges were the main expression of onerous taxes, the ratio can best capture the degree of excessive resource extraction. The mean rate of excess tax is 237.2 with a standard deviation of 231.0.

• (4) Finally and regrettably, systematic information about changing elite network and structure at local levels in rural China for this period (when the literacy rate was very low) simply does not exist. Therefore, lacking data on elite fracturing, we cannot formally analyze its mediating effect on CCP takeover. However, using the methods of Kohler et al. (2011), we can infer the mediating effect indirectly: that is, we compare the overall effects of gentry and militia on CCP takeover with the partial effects of gentry and militia mediated by excess tax to infer the directional effect of elite fracturing on CCP takeover (see the Results section for details). Furthermore, we supplement our quantitative analysis with qualitative historical data, including secondary literature and an in-depth case study of Lianshui based on a variety of primary local history sources.

Table 1 Explanatory factors, measurements, data sources, and corresponding arguments

Table 2 Summary statistics

In addition to these main explanatory variables, we also construct two other variables to use in robustness checks. One of them is the presence of a CCP County Committee prior to 1937, which we use to assess whether the legacy of CCP presence before they were annihilated in the region affected the growth of CCP power during the period of study. We draw data on this variable from county gazetteers. Second, to check if our findings are sensitive to spatial auto-correlation, we compile data on the KMT’s administrative oversight districts (zhengzhi ducha qu)⁶ to cluster standard errors at the district level.

Table 2 presents summary statistics of the dataset and the Appendix, Table 6, details the construction of the variables. Please refer to Harvard Dataverse, https://doi.org/10.7910/DVN/EW81QO, for the replication code and dataset.

Modeling strategy

This study aims to explain Communist control of a county by year, as represented by the presence of a Communist County Committee. As the outcome variable is binary and we have a panel of 93 counties for 8 years, we use logistic regression including county fixed effects and year fixed effects.

Logistic regression with fixed effects nets out the effects of all observed and unobserved time-invariant factors and focuses the analysis on variation in CCP control within counties over time. Specifically, the county fixed effects control for the mean effects of observed and unobserved factors that vary with the county but not with time (i.e., they only have spatial variation) but might affect CCP control. The observed time-invariant factors include (a) elevation and climate (as distinct from weather), (b) the pre-1937 population, farmland, tenancy, taxes, and baojia system (used to compute the Tocqueville-Fei variables of Militia and Excess Tax Rate), and (c) gentry culture legacy. In principle, the factors in (b) did vary with time, but our only source of systematic information, CHMRC, provides data only for the years 1932–1933. So, we treat these as representing structural features of each county and fixed with respect to time during the period of analysis. The unobserved time-invariant factors such as local institutions might be correlated with both CCP control and state penetration (one of the foci of our study) and thus need to be controlled for in the model. The year fixed effects control for factors that are common across all counties but vary with years (i.e., that only have temporal variations) such as the upward trends of both CCP control and Japanese invasion.

With the two-way (county and year) fixed effects included, the identification of the model parameters comes from those factors that vary both with county and year (i.e., that have spatiotemporal variation), such as the Japanese invasion. This means that all variables that do not vary with county and with year cannot be included in the model as stand-alone terms because their main effects are absorbed into the county fixed effects. This is true for the three Tocqueville-Fei variables—Militia, Gentry, and Excess Tax Rate as they only vary with county. In order to estimate their effects, we interact them with Japanese invasion, a variable that varies both with county and year, to reveal their effects in counties and years when they were invaded by the Japanese. For time-varying factors such as CCP leadership in socioeconomic reform and resistance, we include their main effects as well as their interactions with Japanese invasion into the model.

As our data are drawn from historical records, they are, not surprisingly, incomplete. Not much can be done about missing data on CCP control. As for missing data on explanatory variables, one approach would be to limit the analysis to county-years with complete data. However, this would severely reduce the sample and power of the analysis. Instead, following Anderson et al. (1983), we explicitly control for missing data on explanatory variables. This approach is more transparent and preserves the power of the analysis; it also ensures a consistent sample across all estimates and avoids potential sample selection bias.

Furthermore, to guard against spurious correlation due to reverse causation, we specify the explanatory variables with a one-year lag, i.e., in the preceding year. Hence, the outcome variable, CCP control, is for the years 1938–45, while the explanatory variables are for the years 1937–44. Specifically, the logarithm of the odds of the CCP takeover (i.e., the ratio of CCP takeover over no CCP takeover) are:

$$\begin{array}{l}ln\left(\frac{{Prob}({CCP}_{ct})}{1-{Prob}({CCP}_{ct})}\right)=\alpha_0+\beta_1{Japan}_{ct}+\beta_2Militia_c\,{\times\, Japan}_{ct}+\beta_3{MilitiaMiss}_c\,{\times\, Japan}_{ct}+\\\beta_4{Gentry}_c\,{\times\, Japan}_{ct}+\beta_5{GentryMiss}_c\,{\times\, Japan}_{ct}+\beta_6{ExTax}_c\,{\times\, Japan}_{ct}+\\\beta_7{ExTaxMiss}_c\,{\times\, Japan}_{ct}+\alpha_c+\alpha_t.\end{array}$$

(1)

In the above equation, CCP_ct and Japan_ct are, respectively, indicators of CCP takeover and Japanese invasion of county c in year t, Militia_c, Gentry_c, and ExTax_c are, respectively, proportions of local militia and gentry, and excess tax rate in county c, and MilitiaMiss_c, GentryMiss_c, and ExTaxMiss_c are, respectively, indicators of the data being missing in county c, and α_c and α_t are, respectively, fixed effects for county and year.

Results

Conventional explanations

We begin by evaluating the conventional theories. To set the background, Fig. 3 depicts the rising trend of both Japanese invasion and CCP County Committees over the period of study. Following the Japanese invasion of Central China, the number of occupied county seats rose sharply and plateaued, before increasing again in 1944, when the Japanese launched Operation Ichi-Go. In comparison, the number of counties with CCP County Committees rose steadily before dropping in 1942, because of increased KMT-CCP frictions after the New Fourth Army incident and the Japanese scorched earth campaigns (Jinmetsu Sakusen in Japanese) to annihilate Communist bases and “pacify” their rear areas, and gradually recovering thereafter.⁷

Fig. 3

Japanese invasion and CCP county committees: trends

Figure 3 shows that Japanese invasion and CCP control were correlated. However, that correlation might just be coincidental. To examine more rigorously, we turn to regression estimates which include year fixed effects to control for such coincidental correlation. Referring to Table 3, column (1), the coefficient of Japanese invasion, 0.929 (standard error [s.e.] 0.392), is positive and statistically significant. This result is consistent with the consensus view among China scholars that the Japanese invasion paved the way for the Communists (also, Mann, 2012). The estimate implies that, if a county was invaded by the Japanese, the odds that it would be controlled by the CCP in the following year was 2.532 (= exp(0.929)) points higher, which is substantial compared with the average odds of CCP control, 1.638. Indeed, across almost all estimates, the coefficient of Japanese invasion is positive, similar in magnitude, and statistically significant. The magnitude and robustness of this result and its consonance with historical narratives validate our use of statistical methods to analyze the factors for the Communist success.

Table 3 Conventional explanations

In Table 3, columns (2)-(5) report estimates to test theories that emphasized CCP strengths (from here on, for brevity, we write of “CCP control” or “Communist control” as meaning Communist control in the following year). Referring to Table 3, column (2), neither CCP leadership in armed resistance nor military recruitment (“forces”) activities were significantly related to Communist control. Moreover, the coefficients of these factors interacted with Japanese invasion are negative, although statistically insignificant, which is a pattern for all CCP-related factors. By contrast, socioeconomic reforms represented by tax, rent reduction, and land reforms (column (3)), and CCP civilian mobilization represented by establishment and campaigns of CCP-led civilian organizations, such as peasants’ associations, anti-Japanese national salvation associations, and women’s associations (column (4)), were all significantly positively correlated with Communist control in the following year. When combining these factors in Table 3, column (5), CCP socioeconomic reforms and civilian mobilization were positively and significantly correlated with Communist control in the following year.

These results seem to cast doubt on Johnson’s (1962) emphasis on CCP leadership in resisting Japan, suggesting that Communists succeeded not because they won over the local peasantry through their leadership in fighting the invaders. Superficially, the results in Table 3, columns (3)-(5) seem to corroborate Selden’s (1971, also Kataoka, 1974, Wou, 1994) stress on socioeconomic reforms and Chen’s (1986, also Wou, 1994) argument about the CCP’s mobilizational capacity. However, the positive and significant correlations between CCP socioeconomic reforms and mobilization with Communist control in the following year do not imply any causal relationship. They may simply stem from the combination of reverse causality (CCP control in one year leading to socioeconomic reform, for example) and persistence in CCP control (CCP control in one year being correlated with CCP control in the following year), as Fig. 4 illustrates.

Fig. 4

Conventional explanations of CCP strengths: reverse causality

Next, we test the thesis of “social banditry” (Hobsbawm, 1974; Skocpol, 1979; Van Slyke 1986) that topographical conditions explain Communist success. Referring to Table 3, column (6), the coefficient of ruggedness of the terrain interacted with Japanese invasion is negative and marginally significant. That is, if anything, hilly terrain was actually less favorable to Communist takeover than more gentle terrain. This suggests that the thesis of social banditry as “interstitial emergence” (Mann, 1986: 16) may have overstretched an observation that was applicable to an earlier stage of the Chinese Communist movement (1927–1934). At that stage, the CCP established bases in mountainous regions, largely due to inter-ethnic conflict over lands in the peripheries of rural China that they were able to tap into (Averill, 2006; Erbaugh, 1992; Xu, 2018). Our estimates suggest that, as the Communists strived to advance into the core areas of rural society during the Japanese invasion, the importance of these factors faded away.

Table 3, column (7) reports the effects of tenancy rate, which, according to the Sino-Marxist interpretation (Chen, 1933, 1936; China Institute of Pacific Relations, 1938; Paige, 1975; Wolf, 1969) and its neo-Marxist and neo-Weberian adaptations (Moore, 1966; Paige, 1975; Skocpol, 1979; Wolf, 1969), would indicate the inequality of landownership and the extent of class exploitation, thereby increasing the peasantry’s grievances and the chance of revolutionary takeover. Compared with the baseline estimate including only Japanese invasion, the coefficient of Japanese invasion drops by half, as the effect of Japanese invasion on Communist control is concentrated in the counties without data on tenancy. Importantly, the coefficient of tenancy interacted with Japanese invasion is negative but not statistically significant. This does not support the orthodox Marxist and neo-Weberian class-conflict interpretations.

To sum up, Japanese invasion of the county seat is a consistently strong predictor of Communist control in the following year, thus confirming qualitative studies of the role of Japanese invasion in the rise of Chinese Communism, a factor of military power that, as Mann (2012) pointed out, is often overlooked in comparative studies of revolution. Our estimates do not support the emphasis of the “social banditry” thesis on topographical conditions, which appears to be an overstretch from the pattern characteristic of an earlier decade. Nor do we find support for the Chinese Marxist interpretation that tenancy increased rural class conflict and thus Communist success.

The only group of variables that were significantly correlated with Communist control are those measuring CCP strengths, particularly, socioeconomic reforms and mobilization. However, any causal inference from these correlations is challenging. They could be the outcome of reverse causation coupled with persistence in CCP control. These observations raise the question whether, in comparing across counties, the rise of Communist power may have less to do with what they had offered to the peasants, but to pre-existing social structural conditions in the county, in combination with the Japanese invasion, that became receptive to the Communist growth. Since the Tocqueville-Fei perspective points precisely to the dynamic interplay between state centralization and local elite structure that created these local social structural conditions, we now move on to evaluate it.

The Tocqueville-Fei perspective on state centralization and change in local elite structure

Table 4 presents estimates of the Tocqueville-Fei perspective, with column (1) reproducing the estimate considering only Japanese invasion from Table 3, column (1). First, consider the relationship between Communist control and state penetration, as represented by the proportion of militia in the rural population and state resource extraction, as represented by the degree of excess tax. Referring to Table 4, column (2), the estimated coefficient of militia interacted with Japanese invasion is 0.120 (s.e. 0.049)—positive and statistically significant (from here on, for brevity, we do not explicitly mention the interaction of the explanatory variables with Japanese invasion). Subject to the estimate being imprecise, it implies that if state penetration in a county was higher by one standard deviation (13.824),⁸ then the odds of Communist control in the county would be 5.253 (= exp(0.120 × 13.824)) points higher. This is substantial compared with the average odds of CCP control, 1.638.

Table 4 Tocqueville-Fei perspective

The coefficient of excess tax, 0.004 (s.e. 0.002), is positive and marginally significant. It implies that if resource extraction in a county was higher by one standard deviation (168.112), then the odds of Communist control in the county would be 1.959(= exp(0.004 × 168.112)) points higher.

Next, Table 4, column (3) reports an estimate of the effects of gentry culture legacy and state resource extraction. The coefficient of gentry legacy, 0.156 (s.e. 0.080), is positive and marginally significant. The estimate implies that, in the face of Japanese invasion, if gentry culture legacy in a county was higher by one standard deviation (e.g., increased by 23.141), then the odds of Communist control in the county would be 36.966 points higher. This challenges the CCP’s official narrative, which reduced the old gentry to its economic position as a coherent “feudal” landowning class that opposed revolutionary takeover. It thus also calls into question the strand of scholarly literature that takes for granted the notion of a coherent landed class diametrically opposed to the peasant-supported Communist revolution (Skocpol, 1979; Wolf, 1969). The coefficient of excess tax, 0.005 (s.e. 0.002), is positive and statistically significant, and somewhat larger than in the previous estimate, focusing on state penetration and resource extraction.

Table 4, column (4) reports a model to examine the joint effects of state penetration, gentry legacy, and state resource extraction on CCP control. We prefer this estimate as it encompasses all of the factors proposed in the Tocqueville-Fei theory and accounts for possible correlations among those factors. Moreover, the model exhibits better fit on the various conventional measures (Log-likelihood, Pseudo R-squared, and BIC) than the models focusing on the individual explanatory variables separately (Table 4, columns (1)-(3)).

The coefficient of militia is somewhat smaller than in the separate estimate focusing on just militia and excess tax. It implies that a one standard deviation increase in militia would be associated with 4.703 points increase in the odds of CCP control. The estimated coefficient of gentry culture legacy is slightly smaller than in the separate estimate focusing on just gentry legacy and excess tax, and is precise. It implies that one standard deviation increase in gentry legacy would be associated with 26.736 points increase in the odds of CCP control, which is very large. The coefficient of excess tax is the same as in the separate estimates, and implies that one standard deviation increase in excess tax would be associated with 1.959 points increase in the odds of CCP control.

To check whether the results are sensitive to the missing data on militia, gentry, and excess tax, Table 4, column (5) reports an estimate limited to the 55 counties with complete data on militia, gentry, and excess tax. Compared with the estimate including counties with missing data, the coefficients of militia, gentry, and excess tax are similar in magnitude, but less precisely estimated. The loss of precision is somewhat to be expected due to a smaller sample. We prefer the estimate that includes counties with missing data and controlling for such as it is more precise.

Having confirmed the overall effects of state centralization, local gentry culture legacy, and resource extraction on CCP takeover, we proceed to examine the causal mediation mechanisms of the Tocqueville-Fei hypotheses.

To facilitate discussion, we adapt the theoretical model (Fig. 2) to construct Fig. 5, which includes annotation of the causal links as follows: “a” represents the effect of state centralization on resource extraction, “b” represents the effect of gentry on resource extraction, “c” represents the effect of state centralization on elite fracturing, “d” represents the effect of gentry on elite fracturing, “e” represents the effect of resource extraction on CCP takeover, and “f” represents the effect of elite fracturing on CCP takeover.

Fig. 5

Mediation analyses

We first examine the hypotheses regarding state centralization. By Table 4, column (4), the coefficient of excess tax is 0.004 (s.e. 0.002). This is fairly strong evidence that e > 0, which is necessary for hypotheses TF1 and TF3, that state centralization and gentry legacy have affected CCP control through state resource extraction. The coefficient of militia, 0.112 (s.e. 0.042), is positive and significant. However, applying the Kohler et al. (2011) method, as shown in the second row from the bottom of Table 4, the degree to which excess tax mediated the effect of militia on CCP control was small and not statistically significant. This result is not consistent with a > 0, i.e., TF1 that state centralization increased state extraction of local resources, which in turn increased the probability of revolutionary takeover. So, despite state centralization and excessive resource extraction both being strong predictors of CCP takeover, state centralization did not increase CCP takeover through excessive resource extraction. The lack of this mediation effect means that the variation in the magnitude of state extraction of local resources across counties (as measured by the ratio of tax surcharges over regular land taxes) did not reflect the variation in the degree of state penetration in their local communities (as measured by the KMT-organized militia), even though, at the national level, the KMT state centralization was clearly associated with significant increases of state extraction of local resources, as noted by Fei Xiaotong and many of his contemporaries. It also suggests that state centralization affected CCP control through some other channel.

That other channel could be local elite fracturing. But lacking data on that factor, we can only infer indirectly that c > 0 and f > 0. These results would be consistent with TF2, that state centralization increased local elite fracturing, due to the newly empowered state agents, and accentuated local oppression, which in turn, increased the probability of revolutionary takeover.

The coefficient of gentry legacy, 0.142 (s.e. 0.068), is positive and significant. Yet, applying the Kohler et al. (2011) method, as shown in the bottom row of Table 4, the degree to which excess tax mediated the effect of gentry legacy on CCP control was small and not statistically significant. This result is not consistent with b > 0, i.e., TF3 that local gentry legacy reduced state extraction of local resources, and so, reduced the probability of revolutionary takeover. A plausible explanation for this non-effect is that the institution of local assemblies that the gentry leaders dominated and were able to check state power in local communities since the constitutional reform of the early 20^th-century was abolished by the KMT after 1927 when it came to power. This fact also suggests that gentry legacy affected CCP control through some other channel.

That other channel could be elite fracturing. But lacking data on that factor, we can only infer indirectly that d > 0 and f > 0. These results would be consistent with hypothesis TF4 that local gentry legacy increased local elite fracturing, which in turn increased the probability of revolutionary takeover. The fact that old gentry leaders found themselves unable to check the KMT’s excessive extraction of local resources may itself be a motivating factor for their defection to the CCP revolution.

In summary, the empirical evidence largely supports the Tocqueville-Fei perspective on state centralization, gentry legacy, state extraction, and elite fracturing in agrarian communities. While data on elite fracturing are absent, the overall effects of state centralization (indicated by militia), gentry culture legacy (gentry), and state extraction (excessive tax) on the CCP takeover are consistent with the expectations of the Tocqueville-Fei perspective. Yet, our mediation analysis, based on the Kohler et al. (2011) method, does not substantiate the claim, as the TF1 hypothesis has stated, that state centralization increased the probability of the CCP takeover through exacerbating state extraction of local resources, even though both state centralization and state extraction were strong predictors of the CCP takeover. Similarly, the mediation analysis does not substantiate the claim, as the TF3 hypothesis has stated, that gentry culture legacy decreased the probability of the CCP takeover through reducing state extraction of local resources. Nevertheless, while data on elite fracturing are lacking, these results indirectly bolster the claim, as TF2 hypothesis has stated, that state centralization, due to the new state agents that the KMT state building fostered, increased local elite fracturing and thereby increased the probability of the CCP takeover, as well as the claim, as TF4 hypothesis has stated, that gentry culture legacy increased the local elite fracturing and thus increased the probability of the CCP takeover due to the defection of old gentry leaders to the CCP revolution.

Still, absent data on elite fracturing, TF2 and TF4 were evaluated indirectly. Moreover, we could only evaluate each of TF2 and TF4 jointly, and could not separately calculate the parameters, c, d, and f. Hence, we cannot rule out the possibility that c < 0, d < 0, and f < 0, which would also be consistent with the empirical results. That being said, such a statistical possibility (f < 0) would contradict the consensus view in political sociology and social movement scholarship that elite fracturing and division, not elite cohesion, is an important structural opportunity for revolutions and social movements (Bearman, 1993; Foran, 2005; Goldstone, 2001; Goodwin, 2001; McAdam, 1996).

Robustness tests

To check the sensitivity of our findings to the omission of other factors that might confound the effects of state penetration, gentry legacy, state extraction, and elite fracturing on Communist control, Table 5 reports a series of robustness tests, with, for easy reference, column (1) replicating the preferred estimate from Table 4, column (4). First, we check the sensitivity to CCP strengths, with particular attention to two factors that we found to be correlated with Communist control – socioeconomic reform and civilian mobilization. Table 5, column (2) reports an estimate controlling for CCP socioeconomic reform. The coefficient of CCP socioeconomic reform is positive and significant. Yet, the coefficients of militia, gentry legacy, and excess tax are similar to those in the preferred estimate. Apparently, the Tocqueville-Fei factors influenced the likelihood of Communist control independently of Communist efforts to reform society and economy.

Table 5 Robustness checks

Table 5, column (3) reports an estimate controlling for CCP mobilization of the civilian population. The coefficient of CCP mobilization is positive and significant. Yet, the coefficients of militia, gentry legacy, and excess tax are similar to those in the preferred estimate. This result suggests that Tocqueville-Fei factors influenced the likelihood of Communist control independently of Communist activities to mobilize civilians.

As a further check of sensitivity to CCP strengths, we performed an estimate controlling for CCP military leadership. Above, in our investigation of the conventional explanations for the Communist success (Table 3, column (2)), this factor was not significant. Still, it is worth checking whether it might have confounded the effect of the Tocqueville-Fei factors. Referring to Table 5, column (4), it seems that there was no such confound. The coefficients of militia, gentry legacy, and excess tax are similar to those in the preferred estimate.

As a further check of sensitivity to CCP strengths, Table 5, column (5) reports an estimate controlling for whether the Communists had a County Committee in the county prior to 1937. The coefficients of militia, gentry legacy, and excess tax are similar to the preferred estimate without control for historical CCP control. The coefficient of historical CCP control is positive but not statistically significant, suggesting that historical CCP control of a county did not significantly affect CCP success in subsequent years.

The next set of estimates checks the sensitivity of our results to physical and material conditions that might have influenced the likelihood of Communist control. Table 5, column (6) reports an estimate controlling for ruggedness of terrain. The coefficients of militia and gentry are statistically significant and similar to those in the preferred estimate that does not account for ruggedness of terrain.⁹ Of note, the coefficient of ruggedness of terrain is negative. This further confirms our earlier finding that the “social banditry” thesis does not hold. If anything, the Communists were actually less likely to conquer mountainous and remote areas.

Table 5, column (7) reports an estimate controlling for tenancy. The coefficients of militia and excess tax are statistically significant and similar to those in the preferred estimate that does not account for tenancy.¹⁰ Further, the coefficient of tenancy is positive but not statistically significant, suggesting that Communist control was not significantly related to tenancy. This, again, contradicts the Communist claim that a higher tenancy rate reflected stronger class antagonism (Chen, 1933, 1936; China Institute of Pacific Relations, 1938; Paige, 1975; Wolf, 1969), thus leading to more support for the Communist revolution. The finding confirms scholars’ skepticism that the Communist claim seemed a misplaced appropriation of Marxist materialist analysis that grew out of the western European context (Hofheinz 1969: 58; Moore, 1966: 190–91; Russett, 1964: 450–52; Shepherd, 1988). Comparing the null effect of tenancy with the confirmation of the Tocqueville-Fei perspective, we can conclude that the peasant grievances under the KMT rule were not purely economic in the orthodox Marxist sense but political. Besides, their grievances fueled the support for the CCP takeover not only among the peasantry but also among elite defectors like the old gentry leaders, which we will further explicate in the next section.

Lastly, we check whether our findings are sensitive to spatial auto-correlation, by using data on administrative oversight districts (zhengzhi ducha qu), through which the KMT supervised the counties. Table 5, column (8) reports an estimate with standard errors clustered by administrative districts. The coefficients of militia and gentry are positive, statistically significant, and similar to those in the preferred estimate. While the coefficient of excess tax is the same as in the preferred estimate, the standard error is larger, which is to be expected given the higher level of clustering. With due regard for the imprecision of the estimate, we infer that our findings are not sensitive to spatial auto-correlation.

To sum up the robustness checks, the estimated effects of state penetration, gentry legacy, and state extraction are consistent across the various tests and similar to the preferred estimates. It is reasonable to conclude that the findings are robust to the various conventional explanations of the success of the CCP in revolution.

Discussion: making sense of elite fracturing in qualitative analyses

Our statistical analysis has supported, if indirectly, that elite fracturing is an important intermediate process contributing to the growth of Communist power, through two pathways: one is that deeper state penetration increased elite fracturing by boosting abusive new local state agents; the other is that stronger local gentry legacy increased elite fracturing by heightening normative tensions and elite defection from the KMT rule. These two mechanisms of elite fracturing are manifested respectively in two different historical literatures.

On the one hand, Duara’s (1991) acclaimed study of six villages in North China clearly confirms that state penetration led to elite fracturing and decay of local governance. Influenced by Fei’s writing, Duara utilized the data collected by the Japanese-owned South Manchuria Railway Company (Mantetsu) to document a process that he calls “state involution,” through which state penetration led to the drain of old gentry elites and gave rise to an uncontrollable “entrepreneurial brokerage” structure of tax collectors, clerks, middlemen, and bullies. This deterioration of local governance and exacerbation of peasant grievance paved the way for the Communist revolution.

On the other hand, Gregor Benton’s (1999) copious study of the expansion of Communist New Fourth Army guerillas in the Yangtze and Huai Rivers region in 1938–1941 draws attention to the critical role that old gentry leaders played in swinging local support from the KMT to the Communists. While he emphasizes the cultural and social skills that Communist leaders displayed in cultivating local gentry support, it is also clear that the discontent that these gentry leaders harbored toward the KMT corruption made them receptive to the charm offensive of the Communists, whom the KMT decried as ignoble and treacherous.

How did these two different mechanisms interrelate on the ground? Here, we delve into the county of Lianshui, by drawing on a variety of local historical sources related to the county and its first Communist county magistrate Xue Huafu, to illustrate the two mechanisms of elite fracturing, entailed by state centralization on one hand and local gentry legacy on the other.

We choose Lianshui for two reasons. First, it is situated in Northern Jiangsu (one of the three provinces in our statistical analysis) and along the Huai River, an area encompassed by Benton’s (1999) study, although he did not write specifically about Lianshui. Second, Lianshui was the hometown of Gu Zhutong, who, during the war, was commander-in-chief of the KMT’s Third War Zone, whose official jurisdiction encompassed Jiangsu and southern Anhui. The KMT forces, therefore, had a strong reason to defend the county against the CCP takeover.

On the one hand, under KMT rule, Lianshui experienced a high level of state penetration (with two times the average of KMT-organized militia in our sample in 1933). On the other hand, the county had quite a weak gentry legacy. In the period from the end of the Taiping Rebellion to the abolition of the imperial civil service examination, Lianshui produced just one Imperial Scholar, Xue Shangyi, in 1865. Taxation was excessively high in the county (with nearly three times the average of excessive taxation in our sample), thanks to the high level of state penetration and the weakness of its traditional gentry legacy.

In the early 1930s, peasant grievances were severe in the region, creating favorable conditions for the Communists to tap into. Already in August 1930, the Communists staged a major insurrection in six counties in the region, centering around Lianshui county, which alone contributed over 1,000 participants. The insurrection was swiftly crushed by KMT forces within three days, arresting 33 persons, most of whom were executed, including leaders of the local Communist action committee (xingwei) (Gou, 2002, chapter 3). Subsequent minor Communist activities also failed due to deficient leadership and increasing KMT suppression. By early 1935, all Communist cells were rooted out from the county, like elsewhere in the three provinces under our study. Yet, the Japanese invasion of Lianshui in 1939 unleashed the political conditions that state penetration and elite fracturing created for the growth of Communist revolutionary power, a process that the career of Xue Huafu vividly encapsulates.

Xue Huafu (1908–1967) was a scion of Xue Shangyi, the only Imperial Scholar hailing from Lianshui after the Taiping Rebellion (HZWS 1986: 149). He pursued his education in Nanjing (the provincial capital) and Nantong, until a sudden family misfortune—very likely his father’s death¹¹—forced him to quit and return to Lianshui at the age of 22. He first taught in the local school and then ran a cotton mill that closed down after two years. Upon that, he decided to pursue public service and underwent training in the provincial department of civil affairs thanks to the recommendation of his uncle (who was an official in the provincial department of education). He was then appointed the mayor of the second district of Lianshui in 1936. His rectitude, however, put him in tension with the climate of the local administration. He dismissed at once six township chiefs (the level of government between district and village) known for their predation over local peasantry and replaced them with young graduates.¹² This move caused a great uproar among these new elites, who protested to the KMT county government. The county officials were severely divided over this affair. Eventually, he resigned after a mere four months on the job. Later, thanks to his supporters’ lobbying, he was appointed the mayor of another district, an offer he declined, apparently due to his sense of self-pride (Wang, 2008: 51; Lianshui county gazetteer: 968). Nevertheless, he remained a formidable local figure, maintaining his family estate with a well-armed self-defense squad (Wang, 1989: 26).

When the Japanese seized the Lianshui county seat on March 1, 1939, the KMT county government fled to a township seat 40 km away. Xue Huafu contacted two dozens of intellectual youth, among them were Wan Jinpei and Chen Shutong, former Communists of Lianshui origin, who were released from the KMT political prison and returned to Lianshui after the KMT-CCP détente in 1936. Together, they organized a guerrilla force, named the “Lianshui Resistance Independent Battalion,” with Xue as the battalion commander. Meanwhile, the Communists began to send representatives in to rebuild party cells in Lianshui. Seeing that Xue had built a guerilla force, the KMT tried to woo him by appointing him the mayor of the seventh district, the suburban area surrounding the county seat that was under Japanese occupation. Xue accepted the job and commanded his battalion to support the KMT’s attack on the Japanese-occupied county seat in June 1939. Yet, when the KMT attempted to incorporate his battalion in the autumn, he rejected it forcefully (Wang, 2008: 51; Lianshui county gazetteer: 969).

Meanwhile, the Communists began to expand actively in the county and established a working committee in August 1939. With the help of Wan Jinpei and Chen Shutong, who had rebuilt their Communist connections, the Communists finally persuaded Xue to incorporate his guerrilla force into a branch of the Communist-led Eighth Route Army and appoint him as the regiment commander. His defection to the Communists shocked the local KMT establishment, leading to increasingly intensified clashes between his regiment and the remaining KMT forces. The latter even burned down his traditional Chinese courtyard house. On August 27, 1940, as the Communists drove out the remaining KMT forces, took control of the countryside (while the Japanese still occupied the county seat), and established their own government, Xue was elected by an assembly of local notables and Communist representatives to be the first county magistrate and commander of militia under Communist rule. On October 6, Xue led the Communist forces to exterminate the last stronghold of the KMT in the county (Wang, 2008: 52; Lianshui county gazetteer: 32–33, 632; Peng, 2013: 147).

We chose to study Lianshui as it was the hometown of the KMT commander-in-chief in the region and also lay within Benton’s region of study. We do not claim that Lianshui is representative of the counties in our dataset and an expanded qualitative study is beyond the scope of this article. However, by delving deeply into the case of Lianshui, we are able to discern how the patterns that we detect in the statistical analysis actually unfolded in observable events (Steinmetz, 2004). We can especially identify that, consistent with the Tocqueville-Fei perspective, state penetration and gentry legacy both increased elite fracturing: the former by undermining local elite norms and bolstering abusive new state agents and brokers while the latter by deepening old elite alienation and tension with these new agents and brokers. Both heightened the receptiveness to revolutionary takeover. Thus, the two mechanisms of elite fracturing in the existing historical literature both manifested themselves in the illuminating case of Lianshui county and Xue Huafu. And culture played a role both in terms of the legacy of the old gentry’s status norms such as noblesse oblige and in terms of the legitimacy deficiency of KMT state agents and the attending elite fracturing.

Conclusion

As one of the most significant revolutions in modern history, the Chinese Revolution presents an anomaly for the existing theories of revolution. This study builds upon the social geographical approach to revolution pioneered by Markoff (1985; more recently Mazur, 2019, Barrie, 2023) and extends cross-national comparison of countries into a subnational comparison of counties over time. Our analysis indeed shows that Japanese invasion was a consistently significant predictor of Communist control. Besides it, we find weak support for most conventional explanations advanced in the existing literature. On the other hand, our analysis largely substantiates most of the expectations from the Tocqueville-Fei perspective that integrates the threefold focus on state action, elite structure, and cultural change and underscores the dynamic interplay of the centralizing state and local elite structure in shaping a county’s receptiveness to revolutionary takeover. Nonetheless, although state centralization, gentry culture legacy, and state extraction of local resources were all strong predictors for the CCP takeover, our mediation analysis does not support two of the mediating effects, hypothesized by the Tocqueville-Fei perspective, of state centralization and gentry culture legacy through state extraction. The indirect inferences and the qualitative analyses, on the other hand, buttress the effects of state centralization and gentry culture legacy on local elite fracturing, thus leading to revolutionary takeover.

This indicates that the Chinese Revolution was not simply the accumulation of the Communist march from the peripheries in the form of “social banditry” but actually depended on a crucial change in local core institutions in agrarian communities, as a result of changing state-local elite relationship. To put it more bluntly, the Communists succeeded largely because they posed themselves as offering peasants a set of modern repertoires to defend their local and traditional rights that had been violated by a modernizing state and an invading modernized military.¹³

This does not mean that Communist strategies did not matter. The innovative reshuffle of strategy within the CCP from the debacle that threatened its own survival in 1934 to a new paradigm gradually whipped into shape by Mao and his associates in 1936–1938 was instrumental in capitalizing on this structural opportunity. Our point is simple: when we compare counties in a region where the Communists had the same policy platform, it was the state-local elite relationship that explained the varying chance of Communist success. Thus, the common phrase in American electoral politics—that “all politics is local”—has a grain of truth for revolution as well. Of course, it will be far-fetched to claim “all politics is local”: national politics can operate in a relatively autonomous logic from local politics. Nevertheless, our study forcefully shows that local social ecology profoundly shapes the geography of revolution.

One limitation of our analysis is our statistical analysis of the mechanisms of state penetration and local elite structure: these variables need to be interacted with Japanese invasion. Due to data limitations, it is necessary for us to treat the Tocqueville-Fei variables as fixed for each county during the period of analysis. The consequence, to put it simply, is that we can only conclude that the structural condition of the state-society relationship set by state penetration and gentry legacy predicts a significantly higher chance of Communist takeover in face of Japanese aggression. We cannot counterfactually claim that, if Japan had not invaded China at all, the structural condition would have still accumulated to revolutionary change. Nevertheless, in likelihood ratio tests (reported in the Appendix, Table 6), each of the Tocqueville-Fei estimates performs significantly better than the estimate including only the indicator of Japanese invasion, which suggests that the state-local-elite relationship that the Tocqueville-Fei perspective sheds light on makes a significant difference in explaining the variation of revolutionary outcomes.

It is possible that, absent the Japanese invasion, the KMT might have successfully carried out administrative and taxation reforms to alleviate the situation, thus averting the revolutionary crisis. This was likely, especially because by 1934, their military campaigns had forced the Communist movement to the brink of total collapse and the KMT was in a much stronger position to carry out reforms than before. This possibility was foreclosed by the national crisis emerging in 1935–1936 in response to Japan’s increasing aggression in North China, which engulfed the KMT regime before it made a major policy shift toward allying with the Communists to confront the Japanese. The causal conjuncture between foreign invasion and endogenous conditions for revolution is, however, not unique to China. From the French Revolution, the Russian Revolution to Vietnam and Iran, foreign invasions often aggravated domestic structural conditions for revolution. As Fred Halliday once warned: “don’t invade a revolution.” (Goldstone, 2001: 145).

Another limitation of this study is that our research design is sensitive to detecting the structural conditions for revolutionary success in a particular stage of revolution but not in its other stages. Specifically, our study captures the dynamic in the stage of revolutionary growth and expansion when the revolution advances into core institutions in society, but its development is still driven by local factors. This dynamic is different from both the stage of revolutionary emergence and the stage of revolutionary triumph. In the stage of revolutionary emergence, interstitial spaces (Mann, 1986: 16) such as hilly terrains and inter-ethnic tensions are often fertile soils, as the “social bandits” thesis has suggested in the case of the Chinese Revolution. In the stage of revolutionary triumph, the military and material might of revolutionary forces at the national level can shape revolutionary outcome in ways that often override local conditions. While our study focuses on a case in an agrarian society, the same strength on capturing the interaction between state centralization and local elite structure and its attendant limitation to the stage of growth and expansion can be teased out and tested in revolutions since the Iranian Revolution, which were predominantly urban (Arjomand, 1989; Barrie, 2023; Beissinger, 2022; Mazur, 2019).

The relationship between state centralization, elite fracturing, and revolution that this study explicates raises several theoretical and historical implications. First, by extending Goodwin’s (2001) state-constructionist approach, our study goes beyond the narrow Hobbesian assumption that stresses mainly the impact of state predation and extraction on revolution and shows that there are multiple causal pathways that a centralizing state triggers, which can increase the probability of revolutionary takeover.

Second, by subjecting the “groupness” of local elites to critical scrutiny (Brubaker, 2004; Gould, 1996; Tilly, 1978: 62), we challenge the Marxist and neo-Weberian theories of revolution that take local elites as a coherent landed class. Instead, the Tocqueville-Fei perspective suggests that the social formation of local elites is historically variable: it is shaped by state action and strength of collective norms and can in turn shape the outcome of political contention. Revolution is conditioned by elite fracturing and not by elite cohesion. Chinese local elites, on the eve of the Communist takeover, were no more coherent than their French counterparts. The fracturing of local elites between residual gentry leaders and new predatory state agents was characterized by the disintegration of elite collective norms and local legitimacy. Change in local elite structure, in other words, is a culturally conditioned and mediated process. And the endowment of traditional elite status norms in a community is an important factor for how the process unfolds in it.

Third, our analysis not only substantiates Fei Xiaotong’s insights but also suggests an enriched interpretation of Tocqueville’s The Old Regime and the French Revolution. Sociologists of revolution have hitherto mainly focused on his emphasis on state centralization to advance the state-centric perspective (Tilly, 1964; Skocpol, 1979; Markoff, 1985; an exception is Sewell, 1985); whereas cultural historians of revolution have mainly attended to his insights on the “abstract and literary politics” of the Enlightenment (Baker, 1990; Furet, 1981). When Goodwin (2001: 39) proposes to label Tocqueville’s approach as the “state-constructionist perspective,” he appears to call for bridging this divide and expanding the culturalist lens beyond opposition political culture (e.g., the Enlightenment) to state centralization and its consequences. We suspect that one reason why Tocqueville’s analysis of the interplay between state centralization and local elite structure as a cultural process has so far been neglected is that agricultural commercialization and state centralization in France developed in tandem with each other, making it possible for scholars like early Tilly (1964) to treat politics (including state centralization) and culture as secondary and epiphenomenal to economic change.¹⁴ In contrast, China’s “premature” agricultural commercialization (Hung, 2008; Pomeranz, 2000) and late and drastic state centralization temporally separated these two processes, thus accentuating the independent role of politics and culture that Tocqueville has also emphasized. Hence, by highlighting their shared threefold focus on state action, local elite structure, and cultural change, we do not intend to make a simple analogy between the French Revolution and the Chinese Revolution. Instead, our multivariable approach to subnational comparison brings to light their shared threefold focus that can be often lost in cross-national comparison between early twentieth-century China and late eighteenth-century France, due to the significant differences in their relationships to the capitalist world system and in their world-historical contexts. In doing so, this study also calls for an enriched and integrated interpretation of Tocqueville’s classic piece.

Fourth, going beyond France and China, the tension between traditional elite status norms and state centralization that the Tocqueville-Fei perspective sheds light on points to a general predicament that the modernizing state faced in great agrarian empires and states, ranging from Western Europe (Bloch, 1961), to the Ottoman Empire (Barkey, 2008) to Japan (Smith, 1959). Similar systems of the balance of power between the state and traditional local elites existed in these contexts, along with corresponding status cultures. As the modernizing state penetrated local communities and upset these systems of the balance of power, it confronted a great challenge of changing local elite structure while recreating its political authority. Not every modernizing state would provoke a full-blown revolution like the French Old Regime and the Chinese KMT did. Nevertheless, future research can explore how this perspective can help to account for the subnational variations of other forms of political contentions in those modernizing contexts. Notably, in the revolutionary cases where modernizing old regimes had enough capacity to disrupt the old local elite structure but insufficient capacity to contain all the fallouts of this disruption, it was their postrevolutionary successor states that eventually succeeded in turning “peasants into Frenchmen” (Weber, 1976) and extending “the outreach and impact of central power to an unprecedented degree” (Skocpol, 1979: 263), in short, in transforming a status society into a class society (Weber, 1978). In this sense, the postrevolutionary states reaped the fruit of this disruption, since it would have been much harder for them if the old local institutions had not been dismantled. Despite this contrast, there also exists some historical continuity as a result of this local elite fracturing and remaking, which brings us to the fifth and last point.

Fifth, the relationship between local elite fracturing and revolution documented in this study also sheds light on the long-term evolution of state-society in China and helps to make sense of the curious phenomenon that some scholars call “guerrilla capitalism” (Zhang et al., 2013; also, Liu, 1992). Areas in Eastern and Central China liberated by local Communist guerrilla forces in the late 1930s and 1940s were much more likely than those areas liberated by outside Communist forces to develop a prosperous private sector economy in the era of market reform. Our study shows that the Communist success hinged very much on the fact that local guerrilla fighters posed themselves as new protective elites defending peasants’ traditional rights that had been violated by a modernizing state. By drawing attention onto this historical connection, our study helps make sense of how these local cadres became a faction—and often a subordinated faction—within the CCP that looked after local interests and were able to protect grassroots entrepreneurship in Mao’s waning years, which set the foundation for a vibrant private sector in post-socialist China. Xue Huafu, however, did not survive to see that day. He, like many former guerilla leaders with questionable family background, was persecuted during the Cultural Revolution and died in 1967. Thus, like many revolutionaries in the modern era, he was devoured by the revolution he helped make. He was not rehabilitated until the autumn of 1978, when China’s long revolution finally lost its steam and gave way to a new era of Reform (HZWS 1986: 156).

Appendix

Data

Table 6 reports the sources of the data and details of the construction of the variables.

Table 6 Data sources and variable construction

Model fit

Table 7 applies the likelihood ratio test to compare the fits of the models including militia, gentry, and excess tax to that of the simplest model with Japanese invasion as the only explanatory variable (Table 4, column (1)).

Table 7 Model fit