The Case of Hindu-Muslim Conflict in India Part 2

Table 1: Sample Header of Historical Conflict Dataset District (1901) British Province Event Date Attacker Polity Defender Polity Attacker Religion Defender Religion Anantapur Anantapur Anantapur Anantapur Madras Madras Madras Madras Battle Siege Annexation Annexation 1565 1565 N/I 1775 Deccan Sultanate Muhammadans Morari Rao Haidar Ali Rama Raya Vijayanagar Muslims Morari Rao Muslim Muslim Hindu Muslim Hindu Hindu Muslim Hindu Once this material was collected, we did two things. First, we checked to ensure that our research assistants had the same number of conflicts recorded for Anantapur. Second, we cross-referenced these conflicts with secondary source materials to externally validate our data, excluding books and articles also using the IGI as their main source. (ii) Descriptive Analysis of the Data As our dataset allows for identification of conflicts by dyads, we construct two separate categories of historical conflict: first, an index that counts all episodes of conflict, regardless of the religious denomination of the ruler; and second, an index that specifically includes only Hindu-Muslim conflict dyads, which therefore excludes conflict among other combinations, such as colonial wars, or the many battles between independent Muslim rulers that occurred during the medieval era. We include a count of all episodes of conflict as a placebo test, as scholars of Indian history have often noted that Hindu-Muslim conflict has been treated as a monolithic divide without enough concern given to Hindu-Hindu or Muslim-Muslim conflict.83 The underlying accumulation of conflict events in our dataset is shown in Figure 2, as well as in map form in Figure A3 in the Appendix. During the period 1000–1850, Hindu-Muslim conflicts accounted for only a minority of conflicts, and this is reflected in the consistently lower number of incidents at each year in which they are evaluated. This disparity increases sharply during the collapse of the Mughal Empire, and especially in the period after 1757, when colonial wars account for an increasing number of conflicts on the subcontinent. 83 Thapar (1993); Moin (2015) 19 Figure 2: Historical Conflicts, Evaluated by Date: 1000-1850 Notes: Cumulative sum of conflict events (stock) in the dataset, by date; and the annual rate of conflict events (calculated over the century prior to date), by year. To examine the relationship between precolonial conflict and postcolonial ethnic violence, we use the Varshney-Wilkinson dataset of postcolonial Hindu-Muslim riots. This dataset provides a record of all Hindu-Muslim riots reported in the Times of India (Bombay) newspaper for the period 1950 to 1995, including information on the conflict’s precise location.84 Entries were geolocated to India’s post-independence (1951) district boundaries,85 and aggregated for the entire period to produce a count variable for the intensity of Hindu-Muslim violence during the post-independence period in India. The geographic distribution of both historical conflicts along religious lines and contemporary religious violence is shown in Figure 3. 84 85 Varshney and Wilkinson (2006). India’s district boundaries during the first post-independence census of 1951 predate the states reorganization reforms of 1956 and largely preserve colonial-era territorial boundaries. 20 Figure 3: Precolonial and Postcolonial Religious Violence Precolonial Hindu-Muslim Conflict Event Count, 1000-1850 AD. Postcolonial 1950-1995. Hindu-Muslim Riots, Notes: Precolonial Hindu-Muslim conflict is the log count of all conflict events between Hindu and Muslim dyads over the entire series. Postcolonial riots are from the Varshney-Wilkinson database, and show the log count of Hindu-Muslim riot events. Next, in order to examine the relationship between patterns of contemporary ethnic violence and patterns of historical conflict in different periods, we constructed indexes of historical conflict based upon our district-level dataset. These measures are constructed by taking each year and calculating the “stock” of recent historical conflict as sum of past conflicts. Following the precedent of other historical stock variables,86 an annual discount rate on past conflicts is applied during the aggregation procedure, based on the a priori assumption that, in the absence of successive conflicts, the influence of historical conflict will fade over time.87 This also has the benefit of allowing us to better identify how patterns of contemporary violence relate to historical violence in each period. Specifically, the conflict index C for district d at time τ is given by the formula: 86 87 For example, Putterman’s (2014) state antiquity index. A two percent discount rate is applied, ensuring that the magnitude of resentment associated with a single violent event will halve each generation (33 years). 21 Cdτ = Py t=τ (1 + ρ)τ −t · Cdt Where ρ refers to the discount rate, y to the years of the dataset, Cdt to the recorded conflict occurring in a given district in a given year t. We first examine the bivariate relationship between historical Hindu-Muslim conflict and contemporary riots without any controls, taking the year 1200 as our starting point.88 Examining this relationship over the entire historical time-series is a preferable approach to extrapolating from one particular year (as existing work on historical legacies of ethnic violence tends to do) for several reasons. First, it tests whether any statistical association is the result of conflict events that occurred in a single year, or are reflective of a broader historical trend. Second and relatedly, it allow us to identify the events of any historical period as especially covariant with contemporary patterns of religious conflict, as Indian history has cycled through various phases – from the religious violence documented during the early medieval period, to the ecumenicism and tolerance of the early Mughal emperors in the 15th and 16th centuries, to the period of disruption and renewed conflict of the 17th. 88 Conflict over the period 1000-1199 is too sparse to use 1000 as a starting point. 22 Figure 4: Bivariate Correlation between Conflict Indexes and Postcolonial Hindu-Muslim Riots All Conflicts Hindu-Muslim Conflicts Notes: This figure displays the raw correlation coefficients (without controls) between historical conflict stock indexes, evaluated by year (1200-1850), with contemporary (1950-95) Hindu-Muslim riots. Red line indicates statistically significant association at the < 0.05 level. Identical axis scales used in left and right charts. As Figure 4 shows, whereas the bivariate correlation between general historical conflict and contemporary rioting exhibits only trendless fluctation, the bivariate correlation between Hindu-Muslim conflict and contemporary riots becomes positive and significant as the conflict stock approaches the present. Figure 4 also shows that the Hindu-Muslim conflicts of the medieval period are not associated with sites of rioting today. Only beginning roughly in the 18th century does a durable, statistically significant relationship emerge between historical and contemporary religious conflict. (iii) Econometric Analysis Since Figure 4 pinpoints the early modern period (after 1700 A.D.) as historically significant, the rest of our analysis focuses on the years 1500-1850: the periods of Mughal rule, Maratha rebellion, and eventual British supremacy. We estimated 350 separate regressions for each index (1500-1850), again taking each year of the index as the independent measure of conflict, adding a range of control variables in order to estimate the independent association between historical conflict and contemporary Hindu-Muslim 23 violence. Models are estimated in a series of OLS regressions of the form: log(1 + Yd ) = α + log(1 + hdt ) + log(1 + adt ) + β̂1 Xd0 + d Where log(1 + Yd ) refers to the log count (+1) of religious riots in district d during the period from 1950-1995, α to the intercept, hdt to the stock of historical Hindu-Muslim conflict in district d in year t, adt to the stock of all intergroup conflict in district d in year t, β̂1 to a vector of estimated coefficients for the control variables, and Xd0 refers to a matrix of control vectors for historical attributes that could explain variance in religious riots in that district, namely: i) a district-level index of religious fractionalization in 1901, based on denominational data within district territories and cities,89 ii) the percentage Muslim population, by district, in 190190 , iii) a dummy variable for whether, during the colonial era, a district was under direct British rule or governed indirectly as a princely state,91 iv) the land area of the district in square miles, and v) the log district population in 1901. Results are shown in Table 2, taking as independent variables the historical conflict stock as evaluated at 50-year cuts from 1500 to 1850, both for all conflicts, and then only for Hindu-Muslim conflicts. Concurrently, a coefficient plot of the individual index-year coefficients for historical conflict for the full 350-year period are shown in Figure 5, with significant results (at the < 0.05 level) indicated by the red highlighted lines. 89 City and town-level estimates of historical religious denominational numbers are from Jha (2013), supplemented by additional district-level estimates from the IGI. 90 Also from Jha (2013), with additional IGI figures for missing districts and princely states. 91 From Iyer (2010), with additional entries from the IGI for smaller princely states. 24 Table 2: Effect of Historical Hindu-Muslim Conflict Stock. Dependent Variable: Log Count of Hindu-Muslim Riots by District, 1950-95. Conflict indexes evaluated in: 1550 1650 1750 1800 1850 Conflict Stock: Hindu-Muslim -0.511 (0.501) -0.15 (0.437) 0.361 (0.246) 0.618** (0.221) 0.901** (0.316) Conflict Stock: All Conflicts -0.123 (0.154) -0.098 (0.179) -0.001 (0.137) -0.217* (0.103) -0.215* (0.1) Muslim Population, percent of district (1901) -0.175 (0.42) -0.22 (0.427) -0.137 (0.42) -0.098 (0.416) -0.159 (0.414) Religious Fractionalization, by district (1901) 1.004† (0.529) 1.021† (0.529) 0.904† (0.53) 0.925† (0.523) 0.92† (0.521) Indirect or Direct British Rule (0/1) -0.189 (0.136) -0.167 (0.136) -0.174 (0.135) -0.182 (0.134) -0.137 (0.134) Log Population, District (1901) 0.176* (0.068) 0.17* (0.068) 0.166* (0.068) 0.188** (0.069) 0.162* (0.067) Total area (square miles) -0.034* (0.015) -0.032* (0.015) -0.032* (0.015) -0.033* (0.015) -0.027† (0.015) Constant -1.629+ (0.859) -1.574+ (0.862) -1.567+ (0.857) -1.76* (0.862) -1.432† (0.845) 216 0.047 216 0.039 216 0.051 216 0.073 216 0.077 N Adj. r-square Notes: † significant at the 0.1 level; * significant at the 0.05 level; ** significant at the 0.01 level; *** significant at the 0.001 level. Units are 273 districts of colonial India, located within post-independence boundaries. 25 Figure 5: Estimated Coefficients: Historical Conflict to Postcolonial Riots Effect of Conflict Index, by Year: Hindu-Muslim Conflicts Effect of Conflict Index, by Year: All Conflicts Notes: This figure displays the estimated coefficients (with controls) of the effect of historical conflict indexes, evaluated by year (1500-1850), from regression on contemporary (1950-95) Hindu-Muslim riots. Red line indicates statistically significant positive association at the < 0.05 level. (iv) Discussion and Analysis The coefficient for Hindu-Muslim conflict increases in magnitude as the historical conflict index moves toward the present: the stock of Hindu-Muslim political conflict evaluated in 1850 has greater covariance with contemporary religious violence than the index evaluated in 1750, while the index of 1750 has greater covariance with contemporary violence than that evaluated for a century prior. If we assume a latent causal mechanism to link historical conflict with contemporary ethnic violence, then this implies a “half-life” to such effects, in that relatively recent historical events have more causal influence than those occurring in the more distant historical past. This observation points to a discontinuity in the association of historical Hindu-Muslim conflict with conflict in the present day that can be examined more systematically. Before about 1700 there is no positive association between these two variables, yet a sharp increase in the coefficient for historical Hindu-Muslim conflict upon contemporary violence occurs in the late 17th century (+0.7). 26 This observation serves to identify a “critical juncture” in the emergence of Hindu-Muslim tensions in India. During the late 17th century, Aurangzeb broke with the ecumenical policies of his predecessors and implemented a series of repressive policies against Hindus, especially the imposition of the jizyah on non-Muslims. This marked the onset of a sustained period of conflict between the Mughal Empire and local Hindu rulers, most notably Shivaji Bhonsle, who led a rebellion in the Deccan, and whose Maratha polity went on to eclipse the Mughals in size and political influence over the course of the 18th century. The effect of Aurangzeb’s reign is visualized in Figures 6 and A4, which shows a negative association between historical Hindu-Muslim conflict stock and contemporary riots in the year the emperor ascended the throne, yet a significant and positive association after his death. Alternatively stated, districts that experienced Hindu-Muslim conflict during and following Aurangzeb’s reign exhibit significantly higher numbers of Hindu-Muslim riots today. 27 Figure 6: Discontinuity Break in Estimated Effect of Historical HinduMuslim Conflict on Post-Colonial Violence, 1658 (r. Aurangzeb). Notes: Prior to the rule of Aurangzeb in 1658, the estimated association between historical and contemporary Hindu-Muslim conflict was negative and nonsignificant. However, taking the start of Aurangzeb’s reign as a discontinuity break, the average standardized effect of historical conflict rises to +0.172: Alternatively stated, districts of India which participated in uprisings against Aurangzeb and the Maratha campaigns, also experienced significantly higher numbers of Hindu-Muslim riots in the years after independence. We explore this discontinuity more systematically, by comparing the shift in the estimated effects of historical Hindu-Muslim conflict under Emperor Aurangzeb with comparable shifts occurring before and after the reign of each of India’s other principal Mughal rulers. From the start of Mughal rule over India proper in the 16th century to the official end of the Mughal imperium in the 1800s, seven emperors sat on the throne for a duration of two decades or more. Comparing the estimated effect of historical Hindu-Muslim conflict for the decade before and the decade following each emperor’s rule, it is only under Aurangzeb that we find a clear discontinuity: with the average conflict coefficient rising from -0.05 to +0.11 (Table 3). Under earlier Mughal emperors, the effect remained negative 28 and non-significant, a finding that is consistent with the view that violence during this period was more a matter of statecraft than a cause of lasting sectarian grievance. After the reign of Aurangzeb, the pattern of political conflict across India remains consistent with the contemporary geographical pattern of Hindu-Muslim violence, and sees little change under subsequent rulers, barring a further increase a century later during the reign of Shah Jahan III. Table 3: Testing for Discontinuity: Mughal Emperors of India and Shifts in the Estimated Effect of the Historical Conflict Index under their Rule. Emperor Akbar Jahangir Shah Jahan Aurangzeb Muhammed Shah Shah Jahan III Akbar II Reign Conflict Coef. (Start) Conflict Coef. (End) Coef. Shift 1556–1605 1605–1627 1628–1658 1658–1707 1719–1748 1759–1806 1806–1837 -0.06 -0.07 -0.07 -0.05 +0.13 +0.16 +0.26 -0.10 -0.09 -0.03 +0.11 +0.15 +0.27 +0.28 -0.04 -0.02 +0.04 +0.16a +0.02 +0.11 +0.02 Notes: Average standardized coefficient (-1 to +1) for the effect of historical HinduMuslim political conflict on contemporary Hindu-Muslim riots, taken for the ten years before the reign of each emperor, and the ten years following the end of their rule. a Of the seven emperors who ruled lasted for more than two decades, only under emperor Aurangzeb is their a clear discontinuity break: with the average coefficient rising from -0.05 before his rule to +0.11 after. (v) Robustness Checks (a) Placebo tests by conflict dyad. In order to test whether historical Hindu-Muslim conflict is a possible deep determinant of contemporary patterns of religious violence, we have also included a “placebo test” index that uses all conflict events, regardless of the conflict dyad (i.e. also includes “Hindu–Hindu,” “Muslim–Muslim” and colonial conflict events). These coefficients also show a divergence between the estimated effect of “historical 29 conflict” in general vis-a-vis the legacy of historical Hindu-Muslim conflict in particular. Whereas the estimated coefficients for Hindu-Muslim conflict increase in magnitude as we approach the modern era, by contrast, rather than increasing towards the present, the coefficient for “all conflicts” declines from 1500 toward 1650, while the Hindu-Muslim conflict variable remains zero or negative. (b) Further political/historical controls. Of the controls included in Table 2, the estimated association between deep historical conflict and post-colonial ethnic riots remains robust to the inclusion of a variable for British direct or indirect rule. However, as scholars have argued that specific princely states may have exacerbated sectarian tensions as a result of idiosyncratic policies, this may not constitute a fair test.92 A further set of regressions includes fixed effects for five major subgroups of princely state in colonial India the Mysore Kingdom, Hyderabad, Travancore, the remaining Maratha principalities of western and central India, and the Rajput kingdoms located primarily in contemporary Rajasthan (Table A1). Several sub-categories of princely state are associated with a legacy of religious tensions - notably the successors to the Maratha confederacy: though the estimated coefficients for historical religious conflict remain positive and significant. (c) Sensitivity analysis / alternative index aggregation. A further concern may be that our results are sensitive to the method of aggregation, masking important variation between the effects of different types of historical conflict event. We therefore conduct two alternative methods of index calculation as a means of checking the sensitivity of the results to alternative methods of event aggregation. First, our dataset codes a variety of different categories of conflict types that are mentioned in the Imperial Gazetteer, ranging from “annexations” to “battles” to “riots” to “rebellions” and “plunder.” We use these codings to develop two alternative indexes, the first omitting events that lack explicit reference to violent exchange (e.g. annexations), and the second using “political” conflict only (e.g. battles or sieges), omitting categories of a “social” nature (such as riots, assassinations, or rebellions). These are shown in Figure 7(a), and are broadly similar to the results reported using a conflict index that includes all categories of events. Second, we may wish to test the sensitivity of the results to the religious identity of the attacker in the Hindu-Muslim conflict dyad: does it matter if the conflict attacker is Hindu or Muslim? Aggregating separate indices based upon the religious identity of the attacker, we do indeed find an important difference (Figure 7(b)). The index that is aggregated using only events with a Hindu attacker is broadly similar in its association with the 92 See Verghese (2016). 30 geographic distribution of contemporary religious violence as the general index, whereas the index using events with a Muslim attacker has a much weaker association. This implies that, above all, it was districts which participated in the Maratha rebellion and expansion of the late 17th and 18th centuries that today have higher propensity towards Hindu-Muslim violence, rather than being a direct legacy of repressive actions undertaken by Islamic rulers during this period. Figure 7: Sensitivity Analysis: Alternative Conflict Index Estimations (a) Alternate Conflict Event Counts (b) Conflict by Protagonist This pattern of association between Hindu-initiated conflict in the late precolonial period and the pattern of religious riots across India today implies that the Maratha revolt may have played an important role in embedding ethnic tensions in modern India. The estimated coefficients imply that a district without a historical Hindu-Muslim conflict event would experience a +0.63 increase in the number of Hindu-Muslim riots, were it to have had one such event in 1750.93 Alternatively stated, for approximately 2 districts without an experience of historical Hindu-Muslim conflict, the estimated effect of their having a Hindu-Muslim conflict event in 1700 would be for one of these districts to experience an ethnic riot in the post-independence period. Given that the majority of districts in contemporary India (59.9%) have no record of a Hindu-Muslim riot from 1950-1995, this would push such a district into the upper 40 percent of districts in terms of the prevalence of contemporary ethnic violence. 93 This calculation is based on the coefficients from the 1750 regression, assuming a district without a conflict legacy then experienced a single Hindu-Muslim episode in the year 1750. 31 (vi) Explaining Long-Term Persistence: The Marathas and the Birth of Hindu Nationalism Our analysis of historical conflict patterns implies an important association between districts experiencing Hindu-Muslim conflict in the late 17th and 18th centuries and districts prone to ethnic violence in India today: all else equal, Muslims living in districts that were sites of Hindu-Muslim conflict in the century before colonial rule are more likely to be victims of riots during the post-independence period than Muslims living in districts that were not. Disproportionately, such districts are geographically situated in the western and central regions of India, where during the precolonial era the Maratha polity mobilized Hindus against Mughal rule in a century-long territorial struggle. Yet given the long duration between the 18th and 20th centuries, what explains this pattern of persistence? There are several mechanisms that could plausibly connect historical and contemporary violence, such as collective memory or generational transmission. Several scholars, for example, have argued that ethnic groups maintain collective memories of past traumatic events, and that these can reemerge generations later to fuel intergroup conflicts.94 Other studies highlight a mechanism of generational transmission. Acharya, Blackwell, and Sen show that the proportion of the population that was enslaved in 1860 is a strong predictor of racial attitudes in the American south today.95 Lupu and Peisakhin argue that the descendants of deported Crimean Tartars in 1944 display more hostility toward Russians 60 years later.96 Both of these studies find that the mechanisms that enable attitudinal transmission over time are local, including at the level of the family. We suggest an alternative – though not exclusive – mechanism, focusing on the institutional channels through which historical factors persist to the present, and the contingent events that resuscitate ethnic tensions from one generation to another. As many political scientists have noted, Hindu nationalists are a major cause of contemporary religious riots – however, Hindu nationalism is treated as a colonial-era phenomenon. By contrast, we argue that the seeds of Hindu nationalism were sowed in areas of Mughal-Maratha conflict in the 17th century, in the modern-day states of Maharashtra and Gujarat. Historical sources indicate how, after the death 94 95 96 Kapferer (1988); Slyomovics (1998), Eller (1999); Petersen (2002). Acharya, Blackwell, and Sen (2016). Lupu and Peisakhin (2017). 32 of Aurangzeb, the Marathas utilized sectarian violence in their battles with neighboring Muslim states. During the Maratha invasion of Mysore in the late 18th century, for example, Tipu Sultan’s (the ruler of Mysore) court historian Mir Hussein Kirmani describes how “idolators [Hindus] plundered and burned the mosques and houses of Muslims.” 97 If modern-day riots are in fact “pogroms” against Muslims as Brass has suggested,98 then these pogroms began as the Marathas replaced Mughal power in the 18th century. The importance of the Marathas can also be seen in the fact that so many of the key figures in the Hindutva movement are from the formerly Maratha-ruled territories of Maharashtra – for example, Veer Savarkar, K.B. Hedgewar, and M.S. Golwalkar – and why the RSS itself was founded in Maharashtra in 1925. Many of these intellectual leaders of Hindutva were clearly influenced by the historical memory of the Maratha Empire, referencing it in their writings and speeches.99 Another important point to note in explaining how historical conflict persisted in India over time is that precolonial conflicts may have faded by the present day without intervening colonial policies that sustained and reified their existence. We do not disagree with scholars who have noted that the British state emphasized religious divisions; we only disagree with the argument that the colonial state created these divisions. Our historical analysis in this article indicates that identification by religion and communal conflict was already prevalent before colonization; likewise, the British built on top of the inherited religious legacy of precolonial India. V Conclusion In this article, we have sought to reexamine the impact of colonial rule on contemporary ethnic violence. While social scientists have increasingly turned to history to understand contemporary outcomes like democratization and development, this historical turn has been stunted in the study of ethnic conflict. Ethnic violence is treated as a thoroughly modern phenomenon; and, more specifically, ethnic groups and conflict in the nonwestern world are taken as a legacy of European colonialism. Using the influential case of Hindu-Muslim riots in India, we argue for a precolonial perspective: 97 Kirmani (1803). Brass (1997; 2003) 99 A clear example of this is found in the statement to the court made by Nathuram Godse after he was arrested for the assassination of Mohandas Gandhi, arguing that he was continuing “the heroic fight put up by Chhatrapati Shivaji that first checked and eventually destroyed the Muslim tyranny in India.” – Godse and Godse (1949). 98 33 Hindu and Muslim identities began to emerge before British rule, and violence was integral in shaping the construction of these religious communities. We utilize a new dataset of historical conflict in India that spans nine centuries (1000-1850 A.D.), constructed by multiple research assistants from British sources that were then cross-referenced with additional historical material. Using these data, we construct a district-level time series estimate of the stock of communal conflict between Hindu and Muslim polities. We find that religious violence is associated with postcolonial Hindu-Muslim rioting, but that the association begins only with Mughal-Maratha conflict beginning in the late 17th century. We suggest that this result illustrates ethnic counter-mobilization: medieval Hindu-Muslim conflict sowed the seeds of Hindu nationalism, which many political scientists have argued is the proximate cause of contemporary riots. Modern Hindu nationalism was born in the bloody battlefields of 17th century India. While our findings should not be extrapolated to mean that all ethnic conflicts have deep roots, they do suggest that scholars must not ignore precolonial history. The important case of Hindu-Muslim violence in India can only be explained with an eye toward the long durée. 34