Internal wars in history: Triggered by natural disasters or socio-ecological catastrophes?

Abstract

Chaotic events that might be regarded as proximate causes in triggering war have rarely been considered in the large-N quantitative studies of historical warfare. Furthermore, it has not been fully determined which types of chaotic events, natural disasters or socio-ecological catastrophes, are more influential in modulating the likelihood of wars. This study is based on the incidents of 5368 natural disasters, 1478 famines, 5700 epidemics, 456 nomadic invasions, and 1315 internal wars in the agricultural region (including wheat and rice regions) of China in AD 1470–1911, together with Poisson regression and Granger Causality analyses, to explore the catalytic effect of natural disasters and socio-ecological catastrophes in modulating the likelihood of wars in history. The comparison between the wheat and the rice regions is focused. This is the first large-N inter-regional quantitative analysis on this topic. Our statistical results show that, in general, socio-ecological catastrophes are the proximate triggers of internal wars. Specifically, internal wars are triggered by epidemics in the wheat region, and they are ignited by famines in the rice region in historic China. In addition, internal wars in the two agro-ecological zones are revealed to be context-dependent. Also, conceptual models about the synergy of natural disasters and socio-ecological catastrophes in causing internal wars in the wheat and the rice regions are proposed, respectively. The above findings supplement the Malthusian theory by demonstrating the inter-connection among various mortality factors, which has rarely been examined empirically in academia. Moving beyond historic China, researchers are encouraged to boil down war data in other parts of the world by geographic regions in the course of their statistical analysis to examine each region individually in follow-up studies.

Keywords

agrarian societies China history natural disasters socio-ecological catastrophes wars

Introduction

In recent years, there have been an increasing number of large-N quantitative studies measuring the effect of climate change on wars in recent human history. Their general conclusion is that, for pre-industrial societies, wars are attributable to subsistence pressure brought by deteriorating climate such as cooling and drought (Lee et al., 2013, 2016b; Zhang et al., 2005, 2006, 2007a, 2007b, 2011a, 2011b). As agricultural production relied on a low level of technology and, hence, was largely dependent on climate during the time, climate deterioration could result in drastic shrinkage of agricultural production. And the deficit in livelihood resources was aggravated by population expansion engendered by the previous favorable climate. The resulting subsistence pressure either directly causes resource-oriented wars or indirectly causes wars by amplifying existing social tension (Lee and Zhang, 2015; Zhang et al., 2007a). These findings help to illuminate the close connection between the physical environment and human societies in pre-industrial era in a macro (i.e. long-temporal and large-spatial) perspective. As highlighted by renowned philosopher George Santayana (AD 1863–1952), who advised that ‘Those who cannot remember the past are condemned to repeat it’, these findings could also be taken as lessons for contemporary societies about the importance of environmental sustainability, as human societies (especially in less developed countries) remain largely dependent on the physical environment for resources and survival.

Even though the significant connection between climate change and wars in human history has been substantiated with the findings of those large-N quantitative studies, it is worth mentioning that climate change only serves as the root cause or distant trigger of human crisis, in which the effect of climate deterioration is translated to human crisis via climate-driven economic downturn and its associated pathways (Zhang et al., 2011b). Hence, a ~20- to 30-year time gap between the commencement of climate deterioration and the peak of wars is commonly found (Lee et al., 2009a). This may make some observers suspicious about the causal relationship. In addition, the exact timing of the outbreak of wars seems to be better explained by chaotic events such as natural disasters (e.g. flood, drought, etc.) and socio-ecological catastrophes (e.g. famines, epidemics, invasion by perpetrators, etc.). For instance, the Taiping Rebellion, which is believed to be the biggest civilian war in world history, erupted in Guangxi right after a colossal flood in AD 1851. The spread of the rebellion from Guangxi to large parts of southern China was closely related to serious natural disasters in other provinces and a nationwide famine several years before the rebellion (Jiang, 1993). Nevertheless, those chaotic events that might be interpreted as the proximate triggers of war have rarely been considered in the large-N quantitative studies of historic warfare. Out of those studies, emphasis is mainly put on hydro-climatic extremes such as flood and drought (Bai and Kung, 2011; Jia, 2014; Lee et al., 2017b), while the effect of other chaotic events, such as famines and epidemics, has not been measured adequately. In addition, it remains insufficiently explored which kind of chaotic events, natural disasters or socio-ecological catastrophes, are more influential in modulating the likelihood of wars. Given that answers to those questions may help to supplement and advance our understanding about the climate–war nexus, I seek to use big data and statistical analysis to address those questions systematically in this study. This is also the first large-N inter-regional quantitative analysis of the catalytic effect of natural disasters and socio-ecological catastrophes in modulating the likelihood of wars in history.

Data and methodology

Scope of research, study area, and study period

This study seeks to quantify the influence of various chaotic events in inducing wars in history in a macro perspective. The associated findings are mainly for the aggregate of cases, not for individual cases. The agricultural region of China is evidenced by an abundance of war and other related historical documentary records, and sub-national scale empirical research is believed to be the most promising avenue for analyzing the causes and consequences of violent conflicts and mortality factors (Blattman and Miguel, 2010; Dong et al., 2017; Yue and Lee, 2017). Therefore, the agricultural region of China is chosen as the study area (Figure 1). The study period is delimited as AD 1470–1911 (Ming and Qing dynasties), which is commonly covered in all of our datasets (cf. Data). Such spatio-temporal setting sets this research within the context of historic agrarian society. Besides, the study period is largely overlapped with the ‘Little Ice Age’ (c. AD 1400–1900), which can help control the centennial to multi-centennial effect of climate change on wars.

Figure 1.

Wheat and rice regions in China.

Our study area can be further divided into ‘wheat region’ and ‘rice region’ (Figure 1), the two major agro-ecological zones in China. The Yangtze River is the major physical divide of the two zones: the region south of the river is rice-cultivated area, while the region north of the river is wheat-cultivated area. The two regions are characterized by different agricultural practices as well as the psychological differences of people (Talhelm et al., 2014). Such contextual difference between the two regions may result in the regional variation of the sensitivity of internal wars to chaotic events (Lee et al., 2015). As the aggregation of the data of the two regions in statistical analysis may blur the complex dynamics of wars (Lee et al., 2013, 2017a), the two regions will be separately examined in this study.

According to the traditional view of Chinese historians, there are different types of wars, including internal wars, external aggression wars, and others (Chen, 1986). Out of the various types of wars, I focus on internal war in this study because it is an important indicator about social instability throughout Chinese history (Fang et al., 2015). As for chaotic events, natural disasters (floods and droughts), famines, epidemics, and nomadic invasions are events that can trigger wars in ancient China (Lee, 2014; Lee and Zhang, 2010, 2013). Hence, they are taken as the explanatory variables of internal wars, and their effects in triggering internal wars will be measured.

Climatic (e.g. temperature and precipitation) and economic factors (e.g. food prices) are pertinent in explaining the outbreak of wars. Nevertheless, at the inter-annual to multi-decadal time scales, natural disasters, famines, epidemics, and nomadic invasions are also driven by climatic and economic factors (Lee, 2014; Lee and Zhang, 2010, 2013; Pei et al., 2015b), while food prices are also affected by climate variation (Pei et al., 2013, 2014, 2015a, 2016b). The inclusion of those factors as independent variables may incorrectly absorb the signal contained in our concerned chaotic event variables (Hsiang et al., 2013). Therefore, they are not incorporated into my statistical analysis.

Data

Natural disasters are defined as floods and droughts in this study, as they are two of the major natural disasters affecting crop production in historic agrarian China. Their detrimental effect on agricultural production may lead to internal wars (Zhang et al., 2010). The data of floods and droughts are derived from (1) the Yearly Charts of Dryness/Wetness in China for the Last 500-year Period (Chinese National Meteorological Administration, 1981), which contains dryness/wetness grade series for 120 sites in China in AD 1470–1979; and (2) the Yearly Charts of Dryness/Wetness in NW China for the Last 500-year Period (AD 1470–2008) (Bai et al., 2010), which contains the updated dryness/wetness grade series of the 12 sites and the dryness/wetness grade series of seven new sites in northwestern China. In the above yearly charts, a five-point grading system is applied to describe local climatic conditions, ranging from extremely wet to extremely dry (1–5). In this study, the quantification of natural disasters is done by counting the number of sites that have data in any given year whose dryness/wetness grade is 1 (i.e. extremely wet = flood) or 5 (extremely dry = drought) in the rice (Figure 2a) and wheat (Figure 3a) regions, respectively.

Figure 2.

Chaotic events and population density in the wheat region in China: (a) natural disaster, (b) famine, (c) epidemics, (d) nomadic invasion, (e) internal war, and (f) population density.

Figure 3.

Chaotic events and population density in the rice region in China: (a) natural disaster, (b) famine, (c) epidemics, (d) nomadic invasion, (e) internal war, and (f) population density.

Famines are always difficult to be defined objectively. In reference to recent studies (Lee et al., 2016a; Xiao et al., 2015), cannibalism was taken as an indicator of famines. The cannibalism data were obtained from a multi-volume compendium Collection of Meteorological Records in China over the Past Three Thousands Years (Zhang, 2004). Following the practice of Lee et al. (2016a), famine was counted according to the number of counties with cannibalism in a year (Figures 2b and 3b).

Epidemics data were obtained and aggregated from the Collection of Meteorological Records in China over the Past Three Thousands Years (Zhang, 2004), Historical Records of Infectious Diseases in China (Li, 2004), and Epidemic Records in Historical China (Zhang, 2007). The three datasets are tabulated from official dynastic histories and local chronicles. In this study, human endemics and epidemics which were contagious and could potentially spread to a huge population through food, breathing, or human contact within a relatively short period of time were the focus. Those diseases are often recorded as yi or yili (pestilence) in historical documents. Epidemics were counted according to the number of counties affected by epidemics in a year (Figures 2c and 3c), which could better capture the spatial influence and geographic coverage of epidemics (Lee et al., 2016a, 2017a).

For nomadic invasions and internal wars, the data came from a multi-volume compendium Tabulation of Wars in Ancient China (Editorial Committee of Chinese Military History, 1985), which exhaustively records information on the wars in China in 800 BC–AD 1911. This dataset has been employed repeatedly in previous studies (Zhang et al., 2005, 2006, 2007b). In this study, nomadic invasion refers to the invasion of nomadic tribes such as Mongol and Manchu, while internal war denotes peasant rebellions and miscellaneous armed conflicts taking place within the Chinese polities. Both the nomadic invasions (Figures 2d and 3d) and the internal wars (Figures 2e and 3e) are counted in terms of the number of battles.

In the historic period, internal war can be considered as a population density–dependent phenomenon as it is driven by population pressure (Lee and Zhang, 2010, 2013). This is particularly true in late imperial China (Lee, 2014) when there was fourfold increase of Chinese population size brought by the introduction of American food plants (e.g. peanut, sweet potato, and maize) into China (Jiang, 1993). The relationship between chaotic events and internal wars may be mediated by population density, which epitomizes Malthusian pressure on positive checks (Pei et al., 2016a). To calculate population density in the agricultural region of China, I retrieve the historical provincial population estimates in China spanned AD 2–1983 from Chinese Population History (Zhao and Xie, 1988). As the population estimates are at irregular time intervals, the common logarithm of the data points is taken, linearly interpolated, and then anti-logged back to create an annual time series. This method avoids distortions of the population growth rate in data interpolation (Lee et al., 2008, 2009b). In reference to Pei et al. (2016a), the population data are then divided by the area of associated geographic units (i.e. agro-ecological regions) to obtain population density figures (Figures 2f and 3f).

Statistical modeling of internal wars

The association between chaotic events and social responses may be non-linear in the temporal dimension. The application of traditional statistical methods (e.g. linear regression) in examining the climate–society nexus may yield biased estimates (Lee et al., 2017b). Therefore, Poisson regression is employed to examine the inter-annual association between chaotic events and internal wars in the two agro-ecological regions in historic China. Poisson regression is one of the most suitable methods for handling count data (Cameron and Trivedi, 1998). It is designed using the format of a logarithm model (Brouhns et al., 2002). Such method has been applied to scrutinize the influence of climate change on epidemics, mass migrations, and wars in ancient China (Lee et al., 2017b; Pei, 2017; Pei et al., 2015b).

Prior to Poisson regression analysis, collinearity diagnostics of the chaotic events and the internal wars variables in the wheat and the rice regions are run to check whether there is any linear relationship among those variables.

In Poisson regression models, the frequency of internal war is entered as a dependent variable, and the chaotic events, including natural disaster, famine, epidemics, and nomadic invasion as explanatory variables. Population density is entered as one of the control variables. I also add dynasty dummy (dynasty) into the regression models to capture the influence of governance on social instability, in which the period AD 1470–1644 (Ming dynasty) is set as 0, and the period AD 1645–1911 (Qing dynasty) is set as 1. Also, the calendar year and its squared terms (year and year²) are included as control variables to control the time trends of the internal war data (Galloway, 1986; Zhang et al., 2011a). Finally, the frequency of internal wars in the previous 2 years (internal war_t _{− 1} and internal war_t _{− 2}) is entered as control variables to control for auto-correlated errors in the internal war time series (Lee et al., 2013, 2016b).

Although the abovementioned variables were entered into our regression models, the interpretation should remain focused on the effect of the chaotic events on internal wars, with the influence of population density, governance, time trends of internal wars, and auto-correlated errors in time series as controls.

To account for the fact that the social impact of those high-magnitude events may take a few months to materialize, the frequency of internal war in the next year (internal war_t _{+ 1}) was also estimated by the same regression model. Taking the possible heteroskedasticity in data, robust standard error is applied (Greene, 2012).

Apart from measuring the effect of various chaotic events in inducing internal wars, Poisson regression analysis is also run to trace the determining factors of those chaotic events that are shown to be significant in triggering internal wars.

Based on those significant pairs of relationship identified through Poisson regression analysis, conceptual models about the synergy of natural disasters and socio-ecological catastrophes in causing internal wars in the wheat and the rice regions are proposed, respectively. Also, the temporality (i.e. time lag between independent and dependent variables) of those linkages in the models will be further verified by Granger Causality analysis (GCA).

GCA is a method designated in analyzing the relationship of two time series by including their lagged values in bi-variate regression modeling (Granger, 1988, 2001). Such method has been applied to validate the conceptual models pertinent to historic climate–society nexus (Pei et al., 2016a; Zhang et al., 2011b). The carry-over effect of independent variable on dependent variable, which is revealed by their time lag, is statistically identified and compared based on the chosen lag length in GCA. Prior to GCA, Augmented Dickey–Fuller (ADF) test is employed to check the stationarity of the time series. In this study, the GCA lag of the linkages in those models (null hypotheses) is determined according to the concept of ‘proximate trigger’ (cf. Introduction). Hence, the maximum GCA lag of the linkages is set as 2, with the focus put on the carry-over effect within the last 2 years. Akaike’s information criterion (AIC) lag is employed to determine the appropriate lag length. Besides, the GCA results will also be crosschecked by reversed analysis. The temporality of those linkages can be accepted only if the criterion ‘X Granger-causes Y, but Y does not Granger-cause X’ has been fulfilled.

Results

There are 5368 natural disasters, 1478 famines, 5700 epidemics, 456 nomadic invasions, and 1315 internal wars in the agricultural region of China in AD 1470–1911. Descriptive statistics about the data in the wheat and the rice regions are presented respectively in Table 1.

Table 1.

Descriptive statistics of the chaotic event variables.

Variable	Max.	Min.	Mean	Sum	Std dev.
Wheat region
Natural disaster	32	0	5.871	2595	4.754
Famine	199	0	2.701	1194	12.201
Epidemics	122	0	5.425	2398	11.494
Nomadic invasion	38	0	0.269	119	2.022
Internal war	29	0	1.109	490	2.766
Rice region
Natural disaster	25	0	6.274	2773	4.122
Famine	26	0	0.643	284	2.387
Epidemics	76	0	7.471	3302	9.858
Nomadic invasion	22	0	0.762	337	2.599
Internal war	38	0	1.867	825	4.939

For the data in the two agro-ecological regions, the tolerance values and the variance inflation factor (VIF) values of all the independent variables are >0.2 and <5. Hence, no multicollinearity problem is found (De Vaus, 2002) (Table 2).

Table 2.

Collinearity diagnostics of the chaotic event variables.

Variable	Tolerance	VIF
Wheat region
Internal war	0.961	1.04
Natural disaster	0.775	1.29
Famine	0.681	1.47
Epidemics	0.802	1.25
Nomadic invasion	0.988	1.01
Rice region
Internal war	0.985	1.02
Natural disaster	0.848	1.18
Famine	0.725	1.38
Epidemics	0.715	1.40
Nomadic invasion	0.956	1.05

VIF: variance inflation factor.

In the wheat region, both internal war (p < 0.01) and internal war_t _{+ 1} (p < 0.05) are positively correlated with epidemics. The influence of other chaotic events such as natural disaster, famine, and nomadic invasion on internal wars is not statistically significant. In the rice region, internal war is positively correlated with nomadic invasion (p < 0.05), while internal war_t _{+ 1} is associated with famine (p < 0.001) (Table 3).

Table 3.

Estimates of the effect of various chaotic events on internal wars in the wheat and the rice regions in China.

	Wheat region				Rice region
	Internal war		Internal war_{t + 1}		Internal war		Internal war_{t + 1}
	Coefficient (robust SE)	z	Coefficient (robust SE)	z	Coefficient (robust SE)	z	Coefficient (robust SE)	z
Constant	6.088 (18.283)	0.33	11.911 (19.048)	0.63	−1.672 (23.738)	−0.07	−11.514 (22.554)	−0.51
Natural disaster	0.013 (0.022)	0.58	−0.003 (0.022)	−0.12	−0.015 (0.036)	−0.43	−0.010 (0.031)	−0.31
Famine	−0.004 (0.006)	−0.71	0.004 (0.007)	0.55	0.057 (0.033)	1.75	0.086*** (0.024)	3.56
Epidemics	0.016** (0.006)	2.68	0.020* (0.010)	2.00	0.007 (0.010)	0.76	−0.004 (0.010)	−0.43
Nomadic invasion	0.018 (0.024)	0.76	−0.011 (0.018)	−0.61	0.057* (0.027)	2.08	0.038 (0.024)	1.57
Population density	0.007 (0.018)	0.41	0.010 (0.017)	0.60	0.021* (0.009)	2.47	0.026** (0.009)	3.03
Dynasty dummy	−2.217** (0.817)	−2.71	−1.997* (0.811)	−2.46	0.678 (0.660)	1.03	0.696 (0.632)	1.10
Year (10⁻²)	−1.009 (2.163)	−0.47	−1.607 (2.250)	−0.71	0.428 (2.780)	0.15	1.678 (2.683)	0.63
Year (10⁻²)²	0.041 (0.064)	0.63	0.055 (0.066)	0.83	−0.023 (0.083)	−0.28	−0.063 (0.080)	−0.78
Observation	442		441		442		441
Log pseudo-likelihood	−866.814		−823.239		−1405.013		−1395.797
Wald chi²	65.86		78.96		43.53		52.56
Prob. >chi²	0.000		0.000		0.000		0.000
Pseudo R²	0.113		0.157		0.103		0.108

SE: standard error.

All of the data are in annual units. Coefficients and beta are reported, with robust SEs in parentheses.

p < 0.05; **p < 0.01; ***p < 0.001.

However, when the auto-correlated errors in internal war time series are controlled, the picture is a bit different. In the wheat region, epidemics is no longer correlated with internal war_t _{+ 1} (p > 0.05). This makes epidemics and internal war the only statistically significant association there (p < 0.001). In the rice region, internal war_t _{+ 1} remains associated with famine (p < 0.001), while nomadic invasion is no longer significant in triggering internal war (p > 0.05) (Table 4).

Table 4.

Estimates of the effect of various chaotic events on internal wars in the wheat and the rice regions in China, with the auto-correlated errors in the internal war time series controlled.

	Wheat region				Rice region
	Internal war		Internal war_{t + 1}		Internal war		Internal war_{t + 1}
	Coefficient (robust SE)	z	Coefficient (robust SE)	z	Coefficient (robust SE)	z	Coefficient (robust SE)	z
Constant	16.271 (19.799)	0.82	21.329 (20.220)	1.05	−3.541 (27.775)	−0.13	−7.819 (28.181)	−0.28
Natural disaster	0.029 (0.022)	1.35	0.007 (0.023)	0.29	0.008 (0.017)	0.46	0.003 (0.021)	0.15
Famine	−0.004 (0.005)	−0.75	0.005 (0.007)	0.64	0.042 (0.030)	1.42	0.079*** (0.025)	3.18
Epidemics	0.016*** (0.005)	3.22	0.015 (0.010)	1.43	0.003 (0.010)	0.26	−0.011 (0.012)	−0.93
Nomadic invasion	−0.001 (0.024)	−0.03	−0.013 (0.017)	−0.81	0.003 (0.036)	0.07	0.006 (0.025)	0.22
Population density	0.005 (0.017)	0.28	0.008 (0.016)	0.47	0.006 (0.010)	0.66	0.007 (0.010)	0.71
Dynasty dummy	−1.466* (0.737)	−1.99	−1.331 (0.778)	−1.71	0.505 (0.593)	0.85	0.429 (0.524)	0.82
Year (10⁻²)	−2.085 (2.385)	−0.87	−2.618 (2.413)	−1.09	0.542 (3.380)	0.16	1.041 (3.479)	0.30
Year (10⁻²)²	0.066 (0.073)	0.91	0.080 (0.072)	1.10	−0.022 (0.104)	−0.21	−0.036 (0.108)	−0.33
Internal war	–	–	0.078*** (0.018)	4.41	–	–	0.056** (0.018)	3.16
Internal war_{t − 1}	0.101*** (0.018)	5.54	0.054** (0.020)	2.75	0.060*** (0.018)	3.34	0.049** (0.016)	3.11
Internal war_{t − 2}	0.040 (0.021)	1.95	–	–	0.045** (0.016)	2.82	–	–
Observation	440		440		440		440
Log pseudo-likelihood	−790.958		−768.651		−1073.677		−1067.003
Wald chi²	115.75		133.15		340.00		364.70
Prob. >chi²	0.000		0.000		0.000		0.000
Pseudo R²	0.189		0.212		0.313		0.317

SE: standard error.

All of the data are in annual units. Coefficients and beta are reported, with robust SEs in parentheses.

p < 0.05; **p < 0.01; ***p < 0.001.

Given the regression results in which the auto-correlated errors in internal war time series are controlled, internal wars in the wheat region are significantly triggered by epidemics, while internal wars in the rice region are mainly triggered by famines. The same regression procedure is applied to find out the contributing factors of epidemics in the wheat region and the contributing factors of famines in the rice region, respectively. As natural disasters, famines, epidemics, nomadic invasions, and internal wars are inter-related (Lee, 2014; Lee and Zhang, 2010, 2013), when epidemics in the wheat region is examined, natural disaster, famine, nomadic invasion, and internal war are entered as independent variables. When famine in the rice region is examined, natural disaster, epidemics, nomadic invasion, and internal war are entered as independent variables.

In the wheat region, epidemics is associated with natural disaster (p < 0.05), famine (p < 0.05), nomadic invasion (p < 0.001), and internal war (p < 0.01). When the auto-correlated errors in epidemics time series are controlled, epidemics is only correlated with natural disaster (p < 0.001), famine (p < 0.05), and nomadic invasion (p < 0.001). The effect of internal war on epidemics is no longer significant (p > 0.05) (Table 5). Combining this finding with the one in Table 4, it can be further deduced that it epidemics cause internal war in the wheat region, not in reverse.

Table 5.

Estimates of the effect of various chaotic events on epidemics in the wheat region in China.

	Coefficient (robust SE)	z	Coefficient (robust SE)	z
Constant	−31.521 (17.664)	−1.78	−19.183 (17.186)	−1.12
Natural disaster	0.036* (0.015)	2.38	0.045*** (0.013)	3.53
Famine	0.009* (0.004)	2.06	0.008* (0.004)	1.99
Nomadic invasion	0.033*** (0.009)	3.75	0.032*** (0.007)	4.28
Internal war	0.055** (0.021)	2.62	−0.001 (0.026)	−0.03
Population density	−0.002 (0.011)	−0.17	−0.000 (0.011)	−0.01
Dynasty dummy	−1.108*** (0.321)	−3.45	−0.843** (0.322)	−2.62
Year (10⁻²)	3.341 (2.155)	1.55	2.007 (2.092)	0.96
Year (10⁻²)²	−0.080 (0.066)	−1.21	−0.046 (0.064)	−0.71
Epidemics_{t − 1}			0.022*** (0.004)	5.58
Epidemics_{t − 2}			0.002 (0.006)	0.33
Observation	442		440
Log pseudo-likelihood	−2491.845		−2315.597
Wald chi²	112.88		280.31
Prob. >chi²	0.000		0.000
Pseudo R²	0.195		0.250

SE: standard error.

All of the data are in annual units. Coefficients and beta are reported, with robust SEs in parentheses.

p < 0.05; **p < 0.01; ***p < 0.001.

In the rice region, famine is associated with natural disaster (p < 0.05), epidemics (p < 0.001), nomadic invasion (p < 0.001), and internal war (p < 0.001). When the auto-correlated errors in famine time series are controlled, famine is still associated with natural disaster (p < 0.001), epidemics (p < 0.001), nomadic invasion (p < 0.001), and internal war (p < 0.01) (Table 6). Combining this finding with the one in Table 4, it can be found that famine causes internal wart+1, while internal war causes famine, resulting in a feedback loop between famines and internal wars in the rice region.

Table 6.

Estimates of the effect of various chaotic events on famine in the rice region in China.

	Coefficient (robust SE)	z	Coefficient (robust SE)	z
Constant	15.964 (39.092)	0.41	21.517 (37.994)	0.57
Natural disaster	0.071* (0.035)	2.05	0.085*** (0.024)	3.51
Epidemics	0.040*** (0.009)	4.38	0.030*** (0.008)	3.60
Nomadic invasion	0.118*** (0.026)	4.59	0.097*** (0.028)	3.53
Internal war	0.052*** (0.016)	3.31	0.039** (0.015)	2.64
Population density	0.002 (0.014)	0.12	0.015 (0.016)	0.96
Dynasty dummy	−0.527 (0.753)	−0.70	0.211 (0.976)	0.22
Year (10⁻²)	−1.945 (4.636)	−0.42	−2.236 (4.411)	−0.51
Year (10⁻²)²	0.053 (0.138)	0.38	0.047 (0.128)	0.36
Famine_{t − 1}			0.069 (0.038)	1.84
Famine_{t − 2}			0.025 (0.022)	1.15
Observation	442		440
Log pseudo-likelihood	−480.902		−455.254
Wald chi²	116.93		373.26
Prob. >chi²	0.000		0.000
Pseudo R²	0.348		0.382

SE: standard error.

All of the data are in annual units. Coefficients and beta are reported, with robust SEs in parentheses.

p < 0.05; **p < 0.01; ***p < 0.001.

Based on those significant pairs of relationship identified through Poisson regression analysis (Tables 4 –6), conceptual models about the internal wars in the wheat (Figure 4a) and the rice regions (Figure 4b) are proposed. From those diagrams, it can be further generalized that the historical internal wars in the agricultural region in China are mainly caused by socio-ecological catastrophes, while those socio-ecological catastrophes are caused/reinforced by natural disasters or by other socio-ecological catastrophes.

Figure 4.

Conceptual models about the internal wars in the agricultural region in China: (a) wheat region and (b) rice region. Black arrow specifies the significant correlation between two variables identified by Poisson regression analysis; gray arrow indicates the significant temporality between two variables identified by Granger Causality analysis (GCA).

The temporality of those linkages of the models shown in Figure 4 is further verified by GCA. ADF test results show that all of the time-series data pertinent to those models are stationary. Hence, differencing of the time series is not required (Table 7). The GCA lag of the linkages in those models (null hypotheses) is determined by AIC lag (Table 8). Results show that natural disaster and famine Granger-cause epidemics in the wheat region (p < 0.001), not in reverse (p > 0.05) (Table 9 and Figure 4a). This implies that natural disaster and famine may have carry-over effect in triggering the epidemics there. Besides, epidemics Granger-causes internal war (p < 0.001), implying that epidemics outbreak has lasting effect in triggering the outbreak of internal wars. On the other hand, the lagged values of nomadic invasion do not help predict epidemics (p > 0.05), indicating their connection is basically synchronous. In the rice region, however, only nomadic invasion has significant carry-over effect on famine (p < 0.001) (Table 10 and Figure 4b). The linkages between natural disaster and famine, between epidemics and famine, between famine and internal war, and between internal war and famine could not pass the GCA, suggesting the connection between independent and dependent variables in those linkages to be instantaneous.

Table 7.

Augmented Dickey–Fuller (ADF) test on the time series indicated in Figure 4.

Time series	Probability
	No differencing
Wheat region
Natural disaster	0.000
Nomadic invasion	0.000
Epidemics	0.000
Famine	0.000
Internal war	0.002
Rice region
Natural disaster	0.000
Nomadic invasion	0.000
Famine	0.000
Epidemics	0.000
Internal war	0.000

Table 8.

Differencing level and Akaike’s information criterion (AIC) lag of the linkages presented in Figure 4.

Linkage	Differencing level	AIC lag
Wheat region
Natural disaster → Epidemics	No differencing	1
Nomadic invasion → Epidemics	No differencing	2
Famine → Epidemics	No differencing	2
Epidemics → Internal war	No differencing	1
Rice region
Natural disaster → Famine	No differencing	2
Nomadic invasion → Famine	No differencing	2
Epidemics → Famine	No differencing	2
Famine → Internal war	No differencing	2
Internal war → Famine	No differencing	2

Table 9.

Granger Causality analysis (GCA) about the synergy of natural disasters and socio-ecological catastrophes in causing internal wars in the wheat region in China (cf. Figure 4a).

	F
Null hypothesis
Natural disaster does not Granger-cause epidemics	52.177***
Nomadic invasion does not Granger-cause epidemics	2.103
Famine does not Granger-cause epidemics	71.810***
Epidemics does not Granger-cause internal war	43.844***
Reversed analysis for verification
Epidemics does not Granger-cause natural disaster	1.261
Epidemics does not Granger-cause nomadic invasion	2.840
Epidemics does not Granger-cause famine	2.050
Internal war does not Granger-cause epidemics	0.097

All of the data are in annual units.

***

p < 0.001.

Table 10.

Granger Causality analysis (GCA) about the synergy of natural disasters and socio-ecological catastrophes in causing internal wars in the rice region in China (cf. Figure 4b).

	F
Null hypothesis
Natural disaster does not Granger-cause famine	1.207
Nomadic invasion does not Granger-cause famine	8.100***
Epidemics does not Granger-cause famine	2.915
Famine does not Granger-cause internal war	0.565
Internal war does not Granger-cause famine	0.638
Reversed analysis for verification
Famine does not Granger-cause natural disaster	3.087*
Famine does not Granger-cause nomadic invasion	1.545
Famine does not Granger-cause epidemics	22.600***
Internal war does not Granger-cause famine	0.638
Famine does not Granger-cause internal war	0.565

All of the data are in annual units.

p < 0.05; ***p < 0.001.

Discussion

In the Book of Revelation in the New Testament of the Bible, epidemics, wars, famines, and death are described as the Four Horsemen of the Apocalypse. Of the Four Horsemen, war is an intense form of human interaction, which arguably inflicts more suffering on humanity than any other social phenomenon (Blattman and Miguel, 2010). Although previous studies evidence that climate change accounts for the occurrence of wars, the relationship is mainly valid at the decadal or even longer time scales (cf. Introduction). The exact timing of war outbreak is mostly explained by chaotic events (proximate trigger) rather than climate change (distant trigger), even though those chaotic events may have their common root in population pressure and become more frequent in a deteriorating climate (Lee, 2014; Lee and Zhang, 2010, 2013). So far, systematic and empirical examination of the historical impact of natural disasters and other horsemen on wars is sparse.

Based on the big data approach and data-driven method employed in this study, the catalytic effect of some chaotic events in igniting internal wars is found to be statistically significant. Although it is impossible to put chaotic event as the only contributing factor of internal wars, it is generally agreed that when there is any pre-existing acute vulnerability caused by either deteriorating climate or worsening socio-political regime, the impact of chaotic event can be magnified to have devastating consequences on social stability (Kelley et al., 2015).

Internal wars in different agro-ecological zones are triggered by different chaotic events. Although epidemics are usually emphasized as the result of, or at least coincidental to, war in literature (Cooter, 2003; Short, 2010), I find that in the wheat region, epidemics cause internal wars, not in reverse. The possible explanation is that infectious diseases may contribute to wars by altering the balance of power among regions, creating economic and political instability, or increasing social anxiety (Peterson, 2002). Regional human carrying capacity is further eroded, which paves the way for violent conflicts (Butler, 2004). Besides, epidemics affect different groups of people in different ways and to different degrees, which raises the possibility of varying responses by different groups in society. This may bring latent social tensions to the surface, resulting in riot or revolt (Evans, 1988). It is worth noting that the wheat region was traditionally the political center of China. In the study period, except the beginning of the Ming dynasty, the capital city was primarily located in the wheat region (i.e. Beijing). Also, the wheat region was the most populated region in China (Lee et al., 2008), where pastoralists and farmers from different habitats interacted frequently (Pei et al., 2016a). Such context might have increased the contagion of epidemics and their ‘spillover’ effects into the neighboring areas within the wheat region.

Famine has often been emphasized as the trigger of peasant uprising in Chinese history (Chen, 2015; Chu and Lee, 1994; Fang et al., 2015; Ho, 1959). Yet, my findings show that famine is only significant in triggering wars in the rice region of China. This may supplement the existing knowledge about the famine–war nexus in Chinese history. The regionally specific effect of famine on social stability may be attributable to the physical environment setting in the rice region. Geographically, the region is characterized by humid sub-tropical and tropical climate, which is associated with high temperature and abundant rainfall. Even if cooling is severe enough to affect cropping, the more flexible farming system in the south, with its wide range of domesticated species, can adopt alternative crops (Lee et al., 2008). If famine happens in such an agriculturally productive region, it may imply that the food strain there must be too severe for people to cope with. Hence, the associated food strain is translated into internal wars and the reduction of population. On top of the destruction of cultivated land and farmers’ capital goods, the population crash brought by wars greatly reduced the agrarian workforce and left a serious carry-over effect on agricultural production, and consequently the protraction of harvest failure and famines (Chu and Lee, 1994; Zhang et al., 2006). This may result in the feedback loop (i.e. vicious cycle) between famines and internal wars in the rice region.

Thomas Malthus (1798) postulates in his monograph An Essay on Population that global population is destined to exceed the ability of the world’s food subsistence to sustain it. The consequences of this are Malthusian positive checks such as famine, disease, and war. Since Malthus (1798), it has been generally conceived that those positive checks happen when population growth exceeds the growth of resources, which are types of ecological catastrophes. Yet, how those mortality factors interact has rarely been examined empirically in academia. Our findings supplement the Malthusian theory by demonstrating their inter-connection, and also illustrating that internal wars are mainly caused by epidemics and famines (i.e. socio-ecological catastrophes). Socio-ecological catastrophes are shown to be more important than natural disasters (i.e. flood and drought) in increasing the likelihood of internal wars within the context of historic China. It is worth mentioning that the related socio-ecological catastrophes are also caused by natural disasters and by other socio-ecological catastrophes, which is shown in our conceptual models (Figure 4). The simplified pathway could be summarized as natural disasters + other socio-ecological catastrophes → the concerned socio-ecological catastrophes → internal wars. The pathway may give a better idea of how those chaotic events are linked together in different geographic regions. Besides, particular attention should be paid to the temporality of the pathway (Figure 4). The above information may help comprehend the complex mechanism of wars in historic agrarian societies.

Although there is some similarity in the conceptual models of internal wars in the wheat and the rice regions, their differences should not be overlooked (Figure 4), as they imply the internal wars in the two agro-ecological zones to be context-dependent. Subject to differences in their food diversity, ecological vulnerability, and adaptive capacity, the response of human agro-ecological systems to chaotic events is often marked by strong geographic variation (Li et al., 2017). Besides, such differences might have been further reinforced by the contrasting hydro-climatic patterns between the wheat and the rice regions during the ‘Little Ice Age’ (Chen et al., 2015). This further justifies that the two regions should not be aggregated when wars are examined. Researchers who study historic wars are encouraged to boil down data according to geographic regions in the course of statistical analysis and to examine each region individually in follow-up studies (Lee et al., 2013, 2016b, 2017a, 2017b).

Conclusion

I base on the incidents of 5368 natural disasters, 1478 famines, 5700 epidemics, 456 nomadic invasions, and 1315 internal wars in the agricultural region of China in AD 1470–1911, together with Poisson regression analysis and GCA, to explore the catalytic effect of both natural disasters and socio-ecological catastrophes in modulating the likelihood of wars in the two agro-ecological zones (i.e. the wheat and the rice regions) in historic agrarian China. This is the first large-N inter-regional quantitative analysis on this topic. Generally, socio-ecological catastrophes are the proximate triggers of internal wars. Specifically, internal wars are triggered by epidemics in the wheat region, while ignited by famines in the rice region in historic China. Furthermore, internal wars in the two agro-ecological zones are revealed to be context-dependent. Finally, conceptual models about the synergy of natural disasters and socio-ecological catastrophes in causing internal wars in the wheat and the rice regions are proposed, respectively. While the above findings are mainly valid for the aggregate of cases in a macro-historical perspective, not for individual cases in a micro-historical perspective, they supplement the Malthusian theory by demonstrating the inter-connection among various mortality factors, which has rarely been examined empirically in academia. Moving beyond historic China, researchers are encouraged to boil down war data in other parts of the world by geographic regions in the course of statistical analysis and to examine each region individually in follow-up studies.

Footnotes

Funding

This research was supported by the Hui Oi-Chow Trust Fund (201502172003 and 201602172006), Research Grants Council of The Government of the Hong Kong Special Administrative Region of the People’s Republic of China (17610715), and the CAS-SAFEA International Partnership Program for Creative Research Teams. Last but not least, a special thanks to Professor Arlene Rosen and two anonymous reviewers for their valuable comments on the manuscript.

ORCID iD

Harry F Lee

References

Bai

Dong

Zheng

(2010) Yearly Charts of Dryness/Wetness in NW China for the Last 500-Year Period (1470–2008). Beijing: China Meteorological Press.

Bai

Kung

JKS

(2011) Climate shocks and Sino-nomadic conflict. Review of Economics and Statistics 93: 970–981.

Blattman

Miguel

(2010) Civil war. Journal of Economic Literature 48: 3–57.

Brouhns

Denuit

Vermunt

(2002) A Poisson log-bilinear regression approach to the construction of projected lifetables. Insurance: Mathematics and Economics 31: 373–393.

Butler

(2004) Human carrying capacity and human health. PLoS Medicine 1: e55.

Cameron

Trivedi

(1998) Regression Analysis of Count Data. Cambridge: Cambridge University Press.

Chen

(1986) Zhongguo Lidai Tianzai Renhuo Biao [Natural Disasters and Human Catastrophes in Various Dynasties in China]. Shanghai: Xinhua Shudian.

Chen

Feng

et al . (2015) Hydroclimatic changes in China and surroundings during the Medieval Climate Anomaly and Little Ice Age: Spatial patterns and possible mechanisms. Quaternary Science Reviews 107: 98–111.

Chen

(2015) Climate shocks, state capacity, and peasant uprisings in North China during 25–1911 CE. Economica 82: 295–318.

10.

Chinese National Meteorological Administration (1981) Yearly Charts of Dryness/Wetness in China for the Last 500-Year Period. Beijing: Chinese Cartographic Publishing House.

11.

Chu

CYC

Lee

(1994) Famine, revolt, and the dynastic cycle: Population dynamics in historical China. Journal of Population Economics 7: 351–378.

12.

Cooter

(2003) Of war and epidemics: Unnatural couplings, problematic conceptions. Social History of Medicine 16: 283–302.

13.

De Vaus

(2002) Analyzing Social Science Data. London: SAGE.

14.

Dong

Liu

Chen

(2017) Environmental and technological effects on ancient social evolution at different spatial scales. Science China Earth Sciences 60: 2067–2077.

15.

Editorial Committee of Chinese Military History (1985) Zhongguo Junshishi [Tabulation of Wars in Ancient China]. Beijing: Jiefangjun Chubanshe.

16.

Evans

(1988) Epidemics and revolutions: Cholera in nineteenth-century Europe. Past & Present 120: 123–146.

17.

Fang

Yin

et al . (2015) Transmission of climate change impacts from temperature change to grain harvests, famines and peasant uprisings in the historical China. Science in China Earth Sciences 58: 1427–1439.

18.

Galloway

(1986) Long-term fluctuations in climate and population in the preindustrial era. Population and Development Review 12: 1–24.

19.

Granger

CWJ

(1988) Some recent development in a concept of causality. Journal of Econometrics 39: 199–211.

20.

Granger

CWJ

(2001) Essays in Econometrics: Collected Papers of Clive W.J. Granger. Cambridge: Cambridge University Press.

21.

Greene

(2012) Econometric Analysis. Upper Saddle River, NJ: Prentice Hall.

22.

(1959) Studies on the Population of China, 1368–1953. Cambridge, MA: Harvard University Press.

23.

Hsiang

Burke

Miguel

(2013) Quantifying the influence of climate on human conflict. Science 341: 1235367.

24.

Jia

(2014) Weather shocks, sweet potatoes and peasant revolts in historical China. Economic Journal 124: 92–118.

25.

Jiang

(1993) Zhongguo Jindai Renkou Shi [Recent Population History in China]. Hangzhou: Hangzhou University Press.

26.

Kelley

Mohtadi

Cane

et al . (2015) Climate change in the Fertile Crescent and implications of the recent Syrian drought. Proceedings of the National Academy of Sciences of the United States of America 112: 3241–3246.

27.

Lee

(2014) Climate-induced agricultural shrinkage and overpopulation in late imperial China. Climate Research 59: 229–242.

28.

Lee

Zhang

(2010) Changes in climate and secular population cycles in China, 1000 CE to 1911. Climate Research 42: 235–246.

29.

Lee

Zhang

(2013) A tale of two population crises in recent Chinese history. Climatic Change 116: 285–308.

30.

Lee

Zhang

(2015) Quantitative analysis of climate change and human crises in history. In: Kwan

Richardson

Wang

et al . (eds) Space-Time Integration in Geography and GIScience: Research Frontiers in the US and China. Dordrecht: Springer, pp. 235–267.

31.

Lee

Fok

Zhang

(2008) Climatic change and Chinese population growth dynamics over the last millennium. Climatic Change 88: 131–156.

32.

Lee

Fok

Zhang

(2009a) Time lag in the climate change, war, and population relationship: A quantitative analysis. Asian Geographer 26: 83–94.

33.

Lee

Zhang

Fok

(2009b) Temperature, aridity thresholds, and population growth dynamics in China over the last millennium. Climate Research 39: 131–147.

34.

Lee

Fei

Chan

CYS

et al . (2017a) Climate change and epidemics in Chinese history: A multi-scalar analysis. Social Science & Medicine 174: 53–63.

35.

Lee

Zhang

Brecke

et al . (2013) Positive correlation between the North Atlantic Oscillation and violent conflicts in Europe. Climate Research 56: 1–10.

36.

Lee

Zhang

Brecke

et al . (2015) Regional geographic factors mediate the climate-war relationship in Europe. British Journal of Interdisciplinary Studies 2: 1–28.

37.

Lee

Zhang

Pei

et al . (2016a) Demographic impact of climate change on Northwestern China in the late imperial era. Quaternary International 425: 237–247.

38.

Lee

Zhang

Pei

et al . (2016b) Downscaling and disaggregating NAO-conflict nexus in pre-industrial Europe. Chinese Geographical Science 26: 609–622.

39.

Lee

Zhang

Pei

et al . (2017b) Quantitative analysis of the impact of droughts and floods on internal wars in China over the last 500 years. Science China Earth Sciences 60: 2078–2088.

40.

Dodson

Yan

et al . (2017) Quantifying climatic variability in monsoonal northern China over the last 2200 years and its role in driving Chinese dynastic changes. Quaternary Science Reviews 159: 35–46.

41.

(2004) Zhongguo Chuanranbing Shiliao [Historical Records of Infectious Diseases in China]. Beijing: Chemical Industry Press.

42.

Malthus

(1798) An Essay on the Principle of Population. London: The Lawbook Exchange.

43.

Pei

(2017) Migration for survival under natural disasters: A reluctant and passive choice for agriculturalists in historical China. Science China Earth Sciences 60: 2089–2096.

44.

Pei

Zhang

Lee

(2016a) Contextualizing human migration in different agro-ecological zones in ancient China. Quaternary International 426: 65–74.

45.

Pei

Zhang

Lee

et al . (2014) Climate change and macro-economic cycles in pre-industrial Europe. PLoS ONE 9: e88155.

46.

Pei

Zhang

et al . (2015a) Climate change and the macroeconomic structure in pre-industrial Europe: New evidence from wavelet analysis. PLoS ONE 10: e0126480.

47.

Pei

Zhang

et al . (2015b) Epidemics in Ming and Qing China: Impacts of changes of climate and economic well-being. Social Science & Medicine 136–137: 73–80.

48.

Pei

Zhang

et al . (2016b) Temperature and precipitation effects on agrarian economy in late imperial China. Environmental Research Letters 11: 064008.

49.

Pei

Zhang

et al . (2013) Short and long term impacts of climate variations on the agrarian economy in pre-industrial Europe. Climate Research 56: 169–180.

50.

Peterson

(2002) Epidemic disease and national security. Security Studies 12: 43–81.

51.

Short

(2010) War and disease: War epidemics in the nineteenth and twentieth centuries. ADF Health 11: 15–17.

52.

Talhelm

Zhang

Oishi

et al . (2014) Large-scale psychological differences within China explained by Rice versus wheat agriculture. Science 344: 603–608.

53.

Xiao

Fang

Zheng

et al . (2015) Famine, migration and war: Comparison of climate change impacts and social responses in North China between the late Ming and late Qing dynasties. The Holocene 25: 900–910.

54.

Yue

RPH

Lee

(2017) Climate change and plague in European history. Science China Earth Sciences 61: 163–177.

55.

Zhang

(2004) Zhongguo Sanqiannian Qixiang Jilu Zongji [Collection of Meteorological Records in China over the Past Three Thousands Years]. Nanjing: Fenghuang Chubanshe.

56.

Zhang

Jim

Lin

et al . (2005) Climate change, social unrest and dynastic transition in ancient China. Chinese Science Bulletin 50: 137–144.

57.

Zhang

Brecke

Lee

et al . (2007a) Global climate change, war, and population decline in recent human history. Proceedings of the National Academy of Sciences of the United States of America 104: 19214–19219.

58.

Zhang

Jim

Lin

GCS

et al . (2006) Climatic change, wars and dynastic cycles in China over the last millennium. Climatic Change 76: 459–477.

59.

Zhang

Lee

Wang

et al . (2011a) Climate change and large scale human population collapses in the pre-industrial era. Global Ecology and Biogeography 20: 520–531.

60.

Zhang

Lee

Wang

et al . (2011b) The causality analysis of climate change and large-scale human crisis. Proceedings of the National Academy of Sciences of the United States of America 108: 17296–17301.

61.

Zhang

Lee

et al . (2007b) Climate change and war frequency in Eastern China over the last millennium. Human Ecology 35: 403–414.

62.

Zhang

(2007) Zhongguo Gudai Yibing Liuxing Nianbiao [Epidemic Records in Historical China]. Fuzhou: Fujian Science & Technology Publishing.

63.

Zhang

Tian

Cazelles

et al . (2010) Periodic climate cooling enhanced natural disasters and wars in China during AD 10–1900. Proceedings of the Royal Society B: Biological Sciences 277: 3745–3753.

64.

Zhao

Xie

(1988) Zhongguo Renkou Shi [Chinese Population History]. Beijing: Renmin Chubanshe.