PreprintPDF Available

The Electoral Consequences of Mass Religious Events: India's Kumbh Mela

  • May 2023
Authors:
Siddharta Baral
Siddharta Baral
Siddharta Baral
  • This person is not on ResearchGate, or hasn't claimed this research yet.
Gareth Nellis at University of California, San Diego
Gareth Nellis
Michael Weaver at University of British Columbia
Michael Weaver
Preprints and early-stage research may not have been peer reviewed yet.
Download file PDF
Read file
Download citation

Abstract

Mass ritualized gatherings like pilgrimages are central to religious practice globally. Do they generate votes for religious parties? The events may heighten religiosity, enlarging support for parties seen as owning religious policy issues. Such parties might also co-opt the events to organize and campaign. We evaluate the electoral impact of India's Kumbh Mela, a Hindu festival considered the world's biggest human assembly, leveraging its astrologically determined timing combined with districts' proximity by rail to the festival sites. The Kumbh Mela boosts Hindu nationalists' vote share. Mechanisms tests suggest it does so by increasing religious orthodoxy—seen in the adoption of Brahminical dietary practices—and by strengthening Hindu nationalist party infrastructure. Communal violence is unaffected, but the events are electorally polarizing; they cause India's main secular-leaning party to perform better in regions with denser concentrations of religious minorities. Our study offers a new account of how confessional parties make inroads in multiethnic democracies.
Content uploaded by Gareth Nellis
Author content
All content in this area was uploaded by Gareth Nellis on May 31, 2023
Content may be subject to copyright.
The Electoral Consequences of Mass Religious Events:
India’s Kumbh Mela∗
Siddhartha Baral†Gareth Nellis‡Michael Weaver§
May 2023
Abstract
Mass ritualized gatherings like pilgrimages are central to religious practice globally. Do they
generate votes for religious parties? The events may heighten religiosity, enlarging support for
parties seen as owning religious policy issues. Such parties might also co-opt the events to
organize and campaign. We evaluate the electoral impact of India’s Kumbh Mela, a Hindu
festival considered the world’s biggest human assembly, leveraging its astrologically determined
timing combined with districts’ proximity by rail to the festival sites. The Kumbh Mela boosts
Hindu nationalists’ vote share. Mechanisms tests suggest it does so by increasing religious
orthodoxy—seen in the adoption of Brahminical dietary practices—and by strengthening Hindu
nationalist party infrastructure. Communal violence is unaffected, but the events are electorally
polarizing; they cause India’s main secular-leaning party to perform better in regions with
denser concentrations of religious minorities. Our study offers a new account of how confessional
parties make inroads in multiethnic democracies.
∗We are grateful to Nick Gardner, Aura Gonzalez, Toni Liria-Sala, and Jiayi Li for excellent research
assistance. For comments, data, and advice, we thank Poulomi Dhar Chakrabarti, Dave Donaldson, Gabe
De Roche, Karen Ferree, Michael Joseph, Matt Lowe, Ashish Verma, and participants at the 2021 MPSA
and APSA meetings and the 2022 India Politics Workshop.
†Department of Political Science, University of California San Diego. Email: sbaral@ucsd.edu.
‡Corresponding author. Department of Political Science, University of California San Diego. Email:
gnellis@ucsd.edu.
§Department of Political Science, University of British Columbia. Email: michael.weaver@ubc.ca.
How do religious parties win votes? Secular nationalism emerged as the governing ideology of newly
decolonized states in Asia, Africa, and the Middle East in the 1950s and 1960s. Yet its influence has
waned in recent decades. Across a swath of countries—including Turkey, Israel, Indonesia, Sri Lanka, and
Morocco—there has been a surge in support for far-right confessional parties representing majority religious
groups (Juergensmeyer 2019). According to Fukuyama (2018, 66), “[o]ne of the striking characteristics of
global politics [today] is that the dynamic new forces shaping it are nationalist or religious parties and
politicians ... rather than the class-based left-wing parties that were so prominent in the politics of the
twentieth century.” This is particularly true of low- and middle-income settings (Kalyvas 2000). These
trends have thrown cold water on modernization theorists’ claim that secularization and democracy go
hand in hand (Lipset and Rokkan 1967;Fox 2006;Norris and Inglehart 2011). Ethnoreligious nationalism
has emboldened violent extremists and placed minorities at risk (e.g., Tambiah 1992). There is a pressing
need to understand the roots of popular backing for religious parties at a time when liberal democratic
institutions are endangered, maybe as never before, by the exclusionary principles such parties frequently
represent.
Recent academic research spotlights clientelism as a principal tool for religious party expansion—
whether the distribution of cooking oil around election time, or the longer-term provision of basic education
and healthcare to the rural poor (Brooke 2019;Thachil 2014). Less investigated, by contrast, is the role
played by existing religious practices in fostering support. That is the focus of our paper. In particular, we
examine the contribution of mass religious events to the success of religious parties in young, multiethnic
democracies.
Large ritual congregations of adherents are integral to most major religions. The annual Hajj pilgrimage
to Mecca is one of the five pillars of Islam and is considered a sacred duty for all Muslims. Samaritans
journey to Mount Gerizim each year in commemoration of the three Jewish shalosh regalim festivals.
During Holy Week, Catholics from central and northern Mexico convene at the Sanctuary of Atotonilco in
Guanajuato to complete a weeklong cycle of prayers and fasting, ending in a procession. The spectacles
attract the attention of ethnographers, documentary-makers, and historians. Notably absent so far are
detailed assessments of their electoral impacts.
We hypothesize two mechanisms by which mass religious gatherings help religious parties at the ballot
box. One possibility is that joint participation in large group rituals alters people’s sense of their religious
selves, increasing religiosity among marginal believers and reorienting the beliefs of the already-devout.
If greater piety leads individuals to update their political preferences, parties seen as “owning” religious
1
policy issues stand to benefit (Petrocik 1996). Our second explanation holds that religious parties proac-
tively exploit mass religious gatherings to mobilize votes and build their organizations, a strategy we term
platform co-optation. The events have three inherent advantages for religious parties in this regard: (i)
they pre-screen for ideologically sympathetic voters; (ii) provide a safe space where it is hard for hostile
state authorities to regulate party activities; and (iii) supply a focal point for religious parties to coordinate
with civil society affiliates. Taken together, these ideational and pragmatic channels point to mass religious
gatherings having a positive electoral externality for doctrinally aligned parties—an effect transmissible
directly by attendance at the event, and indirectly through spillovers to non-attendees.1
To test these claims, we study India’s Kumbh Mela (“The Festival of the Urn”), a Hindu religious
festival thought to be the world’s largest human assembly. The Kumbh Mela is held at least once every
three years, and rotates between four different locations in northern and western India. The 2013 Mela
in Prayagraj (formerly Allahabad) lasted three months and reportedly hosted 120 million visitors.2The
gatherings lie at the heart of India’s “sacred geography” (Eck 2012). At the festivals, throngs of pilgrims,
sadhus (Hindu holy men), and tourists encamp in vast tent cities, perform devotional practices, listen to
sermons and speeches, and bathe in the salvific rivers.
We measure the causal effect of the Kumbh Mela on local electoral support for Hindu nationalism, a
political project centered on the spiritual and cultural revival of Hinduism, and its protection against alleged
threats from the subcontinent’s other religious traditions, principally Islam (Jaffrelot 1999). Since winning
power at the national level in 2014, the Hindu nationalist Bharatiya Janata Party (BJP) has dominated
Indian politics. It has enacted discriminatory policies targeting the country’s Muslim community.3Partly
as a result, India’s Freedom House rating dropped from “free” to “partly free” in 2021. In the same year,
Varieties of Democracy classed the country as an electoral autocracy for the first time.
We construct a comprehensive dataset that links geo-coded, constituency-level returns for all national
parliamentary elections held since India’s independence in 1947 with a full schedule of Kumbh Melas—39
in total—that we compiled from newspaper reports and secondary sources. Our benchmark analysis uses a
generalized difference-in-differences design. For identification, we exploit variation in a district’s temporal-
1For ease of presentation, this paper uses the term “religious party” to refer to parties that explicitly represent
the religion of the mass religious event in question.
2By comparison, attendance at the Hajj peaked at 3.16 million in 2012.
3See World Report 2021: India, by Human Rights Watch, bit.ly/3qh4uE1. At present, approximately 80 percent
of India’s population identify as Hindu and 14 percent identify as Muslim.
2
spatial distance to the Kumbh Mela at a given election period, a function of the idiosyncratic timing
of Melas relative to national elections, and districts’ proximity via India’s railway network to the four
Mela sites. For tests of theoretical pathways, we use nationally representative sample surveys of consumer
expenditure, census demographics, historical voter surveys, and panel data on Hindu-Muslim violence.
Previewing the results, we document a large positive impact of the Kumbh Mela on the performance of
Hindu nationalist parties. Districts within 450kms of a Kumbh site where a festival has occurred within the
past year experience a 7.6 percentage point increase in Hindu nationalist vote share. The finding is highly
robust. It is supported by placebo tests, and persists across a range of alternative specifications, coding
decisions, and estimation strategies—including those that address shortcomings of the two-way fixed effects
estimator (De Chaisemartin and d’Haultfoeuille 2020). The Mela’s effects are especially pronounced in the
earlier phase of Hindu nationalist-party growth. Capitalizing on existing religious events may become less
electorally valuable as religious parties become more integrated into mainstream politics.
In terms of mechanisms, there is evidence that the Kumbh Mela promotes orthodox Hindu religiosity,
seen in the increased take up of the upper-caste Hindu practice of vegetarianism. The results thus lend
credence to the claim that social identity change lies behind mass religious gatherings’ impact on voting
behavior. We also substantiate our second argument: that mass religious events offer religious parties
a readymade platform for organizing and mobilizing. Qualitatively, there are direct accounts of Hindu
nationalists piggybacking on the festivals to attend to internal party business and raise party visibility.
In a case study using two of the earliest national voter surveys fielded in India, we demonstrate that
the Prayagraj Kumbh Mela of 1971 strengthened Hindu nationalist party infrastructure nearby. We end
by examining the Kumbh Mela’s wider social and political impacts. While the events do not exacerbate
communal violence, they are electorally polarizing. In districts with a higher proportion of Muslims, India’s
principal secular-leaning party gains votes in response to the Melas, whereas other parties lose ground.
Scholars have underscored the need for greater focus on religion’s interaction with democratic processes,
particularly in non-Western contexts (Woodberry 2012;Chhibber and Shastri 2014). Heeding this call,
we tie mass religious observance to voting in the world’s largest electorate. In doing so, we add to the
specific study of pilgrimages’ attitudinal and behavioral impacts. Clingingsmith et al. (2009) find that Hajj
attendance increases conformity with global Islamic practices, and improves attitudes toward ethnic and
religious outgroups. Christia et al. (2019) show that religious socialization during the Ashura pilgrimage to
Karbala, Iraq produces a convergence in sectarian norms. We explore the electoral consequences of mass
religious events for the first time—shedding light on the macro-level political shifts they bring about.
3
Second, our research dovetails with work looking at religion’s social-psychological implications for
politics. Speeches with religious content influence the nature and extent of participation in ethnically
divided democracies (McClendon and Riedl 2019). Catholic clergy have been politically influential in
Brazil (Tu˜n´on 2017). Islamist parties enjoy an electoral head start owing to their perceived sacred links
(Grewal et al. 2019). We show that mass religious gatherings—where sermonizing and religious iconography
are prevalent—can affect social identities, and intensify political polarization.
Finally, we expand the literature on the political activation of social cleavages (Lipset and Rokkan
1967). Without denying the significance of the materialist tactics central to recent research, we show
how religious nationalists can harness longstanding religious practices to reap electoral rewards. The
repurposing of supposedly apolitical religious events in service of electoral competition is a notable example
of the “modernity of tradition” (Rudolph and Rudolph 1984).
Argument
Why might large religious gatherings help religious parties pick up votes in elections? This section lays out
two reasons: one focused on identity change, the other on platform co-optation. It rounds off by discussing
how these mechanisms apply to both event participants and non-participants.
Religiosity and identity change
A mass religious gathering could increase the religiosity of attendees, and the importance they assign
to religion in their everyday lives. Identity change at the events may come about through psychological
and informational channels. Group ritual activities lead to “depersonalization” and in-group cohesion,
according to processual models in sociology (Olaveson 2001). Durkheim (2008, 162) famously portrayed
ritualized collective interactions as societal self-worship, marked by “a kind of electricity that quickly
transports [participants] to an extraordinary degree of exaltation.” In this state of collective effervescence,
he argued, individualism breaks down, prompting participants to bind more strongly with the symbols
and value-systems of the community. Turner outlined a similar phenomenon, communitas, whereby col-
lective, rhythmic performance in a sacred, “liminal” setting fuses once-atomistic social structures into “a
homogenous, undifferentiated whole” (Turner 1977, 177).
Beyond psychological effects, collective rituals—which pilgrimages and processions epitomize—are venues
where higher-order beliefs are forged, providing a rationalist explanation for identity switching in these
contexts. Chwe (2013) presents a formal model in which rituals foster common knowledge by permitting
community members to observe each other’s behaviors and attitudes. Rituals not only communicate what
4
community members’ beliefs are, they make clear that other community members know about the existence
of each others’ beliefs: “I know that you know ... that I know,” and so forth. Gatherings with ritualistic
elements thus solve coordination problems, driving group-level changes in practices and norms.
Increased religiosity caused by mass religious gatherings may be an important input into political
behavior, but it is not enough by itself to move election outcomes. For that to occur, two further conditions
are necessary. First, more religious citizens must go on to demand a bigger role for religion in state policy-
making. Such a stance may follow from the content of certain belief-systems, from the instrumentalization
of religious wedge issues by election-minded elites, or from religion being at the top of voters’ minds while
elections are ongoing (Nelson 2011;Wilkinson 2006;Zaller 1992). Second, on the supply side, there must
be at least one party or candidate in a position to respond to these demands; i.e., there needs to be an
outlet for devout voters to make their preferences heard.
Platform co-optation
The previous argument presents confessional parties as mostly passive beneficiaries of large religious events:
gatherings mint more religious voters, who then cast their votes for religious parties. By contrast, our other
explanation emphasizes the deliberate actions of parties themselves. We propose that religious parties
engage in platform co-optation, taking advantage of mass religious gatherings to conduct on-site outreach
to potential voters, and to enhance the effectiveness of their organizations.
Three features of popular religious gatherings make them ideal for co-optation. First, the events screen
in large numbers of individuals and groups likely to be sympathetic to religious parties’ cause. Tracking
down persuadable voters is an uphill battle for most parties, which is one reason why partisan persuasion
efforts often fall flat (Kalla and Broockman 2018). Mass religious gatherings mitigate the search problem
owing to participant self-selection. The events attract not only pious voters but also those marginally
inclined toward religion (e.g., the pilgrim-vacationers at the Tabbard Inn in Chaucer’s Tale of Beryn). It is
therefore reasonable to expect that attendees will be more receptive to religiously couched political appeals
than the average citizen. By the same token, the events are fertile grounds for recruiting the committed
rank-and-file cadres needed to staff local party cells.
Second, mass religious gatherings provide a “safe space” where religious party activities are shielded
from sometimes hostile state authorities. Democracies and autocracies have, at times, tried to restrict
public religious practices. But regulating religious spaces invites blowback in deeply religious countries.
Full-scale suppression of collective worship has generally proven impossible. Examples of political move-
ments using religious spaces for officially proscribed activism include the Iranian revolution, when mosques
5
“served as centers for dissent, political organization, agitation, and sanctuary” (Esposito 1999, 110). Re-
ligious pilgrimages were revived at the end of communist rule in Slovakia, with the Christian Democratic
Movement immediately embracing their “political potential” (Doellinger 2002, 226). Well-known, too, is
the pivotal part Black Baptist churches played in the civil rights movement in the southern United States.
Analogously, mass religious events can offer safe-haven to confessional parties trying to sign up voters and
volunteers.
The events alleviate a third challenge faced by religious parties: their need to coordinate activities
with regional branches and civil society affiliates (Wickham 2015). Religious parties at their early stages
of development commonly find themselves geographically over-stretched, resource-constrained, and reliant
on outside groups for manpower and ideological direction. Mass religious gatherings are a natural point of
convergence for handling organizational affairs.
For these reasons, then, we should expect mass religious events to be a boon to religious parties, and
not to their competitors. But the opportunities should not be overstated. Attempting to get electoral
mileage from religious festivals involves risks. If religious parties are seen as tarnishing a sacred occasion
by injecting it with politics, any vote gains could be offset by public anger. Savvy party operatives must
devise strategies for engaging with religious attendees that avoid these pitfalls.
Spillovers
We expect mass religious gatherings not only to shape the political preferences of event participants; they
might also mold the voting behavior of a wider segment of the voting population, above all in geographically
proximate areas. Participants induced to support religious parties by going on pilgrimage may transmit
their new preferences to their families and close social networks. Organizational improvements resulting
from the gatherings should prove a significant multiplier, expanding the capacity of religious parties to
secure votes well beyond the boundaries of the event itself. Local media coverage could also profit religious
parties, giving them visibility and credibility—by dint of their association with the event—that they would
not otherwise enjoy. The “treatment” is bundled, therefore. Furthermore, the festivals’ impacts should
resemble a ripple pattern, being greatest in the vicinity of the mass gathering and in the time period shortly
after its conclusion, then decaying as both distance and time from the event grow.
Background
The Kumbh Mela. The Kumbh Mela is a Hindu religious gathering held approximately once every
three years across four Indian cities: Haridwar and Prayagraj in the north of the country, and Ujjain
6
A B
Figure 1: Panel A: Map of India showing the locations of the four Kumbh Mela sites and the average vote share
received by Hindu nationalist parties across all Lok Sabha elections, 1951–2019, at the 1961 administrative district
level. Panel B: Timing of the Kumbh Melas—including the Ardh melas—across the four event sites (red circles)
relative to India’s Lok Sabha elections (vertical blue dashed lines).
and Nashik in the center-west (see Figure 1A). Haridwar and Prayagraj also host Ardh (“half”) Kumbh
Melas midway between the full Kumbhs. The timing of the Melas is determined astrologically, based on the
positions of Jupiter, the sun, and the moon.4During the festival, which generally lasts two to three months,
pilgrims seeking spiritual purification bathe in the rivers at the sites. On the most auspicious days, the
primary bathing spots are reserved for Hindu sadhus who perform the shahi snan: ritual submersion at the
end of elaborate processions, during which groups of ascetics (akharas) brandish ceremonial weapons and
parade their leaders on gold and silver palanquins in front of large crowds. The Kumbh itself exemplifies
a Durkheimian effervescent environment: “The entire atmosphere is saturated with the religious fervor of
chiming bells, incense, flower fragrances, Vedic hymns, mantras, and the beating of drums” (Bhela 2010,
100).
The origins of the Kumbh Mela are unclear. Some ascribe it to Adi Shankaracharya, an eighth-century
philosopher who saw regular meetings of Hindu scholars and priests as a means of strengthening Hinduism
against Buddhism (Lochtefeld 2004). Violence marred the Melas in the precolonial period, as competing
akharas vied for status and royal patronage. The Melas also became important commercial fairs at that
4See Online Appendix Ofor further details, and Online Appendix Pfor a discussion of a rare dispute over Mela
timing.
7
time. The expansion of the railway network during British rule led to ballooning attendance in the
nineteenth century (Maclean 2008). Today, regional governments are tasked with organizing the festivals,
erecting “pop-up mega-cities” replete with campgrounds, public health facilities, sanitation, and extensive
security (Khanna et al. 2013). Trains—the principal mode of transport for poorer Indians—are specially
commissioned to convey pilgrims and tourists. The Uttar Pradesh state government reportedly spent USD
500 million on the 2019 Prayagraj Kumbh Mela, which is estimated to have employed 300,000 people.5
Hindu nationalism. Politically, the 76 years since India’s independence have seen the decline of secular
nationalism—embodied in the Congress party—and its replacement by Hindutva, a majoritarian worldview
that took root in the 1920s and is today represented by a family of organizations known as the Sangh
Parivar.6Several political parties have claimed to speak for the “Hindu nation.” Of them, the BJP has
made the biggest electoral strides. After its founding in 1980, the BJP seized on divisive socio-political
issues. It accused the then-dominant Congress party of “pseudo-secularism” and pandering to India’s
Muslim minority. Its pro-market economic policies appealed to India’s burgeoning middle class. It has also
won the support of upper-caste Hindus opposed to affirmative action policies for backward-caste groups.
The BJP held India’s prime ministership between 1998 and 2004 at the head of coalition governments.
Under Narendra Modi, it first achieved its own legislative majority in India’s lower house in 2014.
Kumbh Melas as political sites. The political significance of the Kumbh Mela has not been lost on
historians and journalists. The Hindu Mahasabha—an early pressure group advocating for Hindu unity—
was initially convened at the 1915 Kumbh Mela in Haridwar (Bapu 2013). During the anti-colonial struggle,
the nationalist movement disseminated its messages of swaraj,swadeshi, and satyagraha at the gatherings,
while association with the Mela helped imbue it with “a sense of divine right” (Gould 2004, 85). The
festival has remained a political hotbed in the democratic era. Parties set up elaborate camps at the
Kumbhs, and top politicians make highly publicized appearances to reinforce their religious credentials.
Despite the projection of the Kumbh Meha as an inclusive, pan-Indian cultural event, in recent decades
Hindu nationalist organizations have disproportionately conscripted it to advance their political agenda.
At the forefront has been the Vishwa Hindu Parishad (VHP), a militant Hindu nationalist organization
associated with the Rashtriya Swayamsevak Sangh (RSS). It assembled the inaugural World Hindu Con-
5“Kumbh Mela 2019 to Generate Revenue of Rs 1.2 Lakh Crore: CII,” Business Today, January 21, 2019,
bit.ly/2PXjCc7.
6Although the Congress has a reputation for being broadly secularist, in practice it has, at times and at different
levels, played on religious sentiments for political gain (Sherman 2022, ch. 3).
8
ference at the 1966 Kumbh in Prayagraj.7At the 2016 Kumbh in Ujjain, the BJP state government put
on a Vaicharik Kumbh, or “Kumbh of thoughts,” that brought together Hindu nationalist politicians and
religious leaders.8In short, hardline groups have repeatedly used the event to elevate the Hindutva cause.
India’s pilgrims. To what extent do pilgrims resemble the rest of India’s population? Knowing this
can help us understand who might be politically impacted by the Kumbh Mela. Accounts indicate that
pilgrimage is a popular activity and not the preserve of “true believers” (e.g., Eck 2012, 64). Online
Appendix Figure A2 analyzes nationally representative Pew data collected in 2019–20. It finds the demo-
graphic, behavioral, and attitudinal differences between Hindu pilgrims and non-pilgrims to be modest in
magnitude. For instance, only 44 percent of Hindu pilgrims report having voted for the BJP in the 2019
national election, compared to 41 percent of non-pilgrims, showing that the pilgrim population encom-
passes more than staunch Hindu nationalists. In a survey of attendees at the 2016 Ujjain Kumbh Mela,
the most commonly cited reasons for attending the event turned out to be non-religious ones (see Online
Appendix Figure A3). Surveys thus point to there being a large pool of pilgrims potentially convertible to
political Hindu nationalism.
Methods
We describe our empirical strategy for identifying the impact of the Kumbh Mela on electoral outcomes.
The core of our approach lies in measuring how politics and society in a locality are affected by (i) ease-
of-access to the Kumbh Mela festivals combined with (ii) the recency of festivals at each site.9
Data and coding
The unit of analysis for the primary tests is the 1961 administrative district during a given Lok Sabha
election.10 For the main results, we focus on a balanced panel of districts observed across 17 election
7“All Set for the Kumbh Mela: Steps to Regulate Pilgrim Traffic,” Times of India, January 7, 1966,
bit.ly/3wIOoWM.
8“Congress Alleges Scam in Ujjain,” Hindu, July 19, 2016, bit.ly/3wRBwxu.
9Online Appendix Nprovides evidence from cellphone records that geographic proximity to the Kumbh is pre-
dictive of attendance.
10Excluded from the sample are the sparsely populated Andaman Islands, the archipelago of Lakshadweep, and
Daman and Diu. These territories are either wholly or partly not on the Indian mainland and are thus inaccessible
to the Kumbh Mela by rail.
9
years.11
Treatment variable. We draw on multiple sources to construct the right-hand-side data. First, we use
the online archive of the Bombay edition of the Times of India, India’s newspaper of record, in addition to
assorted historical and astrological documents, to produce a complete schedule of the start and end dates
of all full and half Kumbh Melas held in Prayagraj, Ujjain, Haridwar, and Nashik dating back to 1943.12
The timeline of Kumbh Melas is displayed in Figure 1B, and a detailed listing of data sources for each
event is given in Supplementary Information Table S2. Second, we append the start-dates of each national
election cycle to the dataset (taken from Agarwal et al. 2021). Third, we pinpoint the latitude/longitude
coordinates of the four Kumbh Mela grounds. Finally, we use geodata on the Indian rail network as it
existed in 1956 (digitized by Donaldson 2018) to create a spatial network linking all 1961 district centroids
to all four Kumbh sites by train.13 For each district dduring election t, we then code a binary indicator
as follows:
Recent nearby Kumbhdt =
1,if a Kumbh has occurred within 450kms by rail within the past 365 days
0,otherwise
(1)
The choice of distance and time cutoffs involves some arbitrariness. The distance cutoff represents 12 hours’
train-travel time,14 while our prior was that any localized Kumbh effects—the product of a months-long
festival—would endure at least a year. We demonstrate that the estimated impacts are not sensitive to
these particular choices.
11See Supplementary Information Dfor panelView visualizations of data structure and coverage. We present
robustness tests below demonstrating that the results are virtually unchanged using the full, imbalanced panel.
12We include the Ardh Melas because they have attained near-equal prominence to the full Kumbhs.
13We use Dijkstra’s algorithm to compute the shortest railroad distance from each district centroid to each of the
four Kumbh sites. We employ a map of the network as it existed close to the start of our study period to avoid
post-treatment bias, which could arise if rail construction occurred in response to political dynamics induced by the
Kumbh Mela. In practice, there has been limited rail expansion since the 1950s; our calculation shows that the 1956
network stood at 84 percent of its total 2001 length. Online Appendix Gprovides a robustness check on the main
results using the 2001 rail map.
14The average Indian train moves at 38kms per hour.
10
Outcomes. We gathered constituency-level electoral returns for all national (Lok Sabha) elections, from
the first general election in 1951-2 up to the most recent one in 2019. We then identified the geographic
locations of every Lok Sabha constituency that has existed since independence. Constituency boundaries
are periodically redrawn to correct for malapportionment. To preserve stable geographic measurement
units over time—needed for the panel analysis—we assign every constituency to the 1961 district to which
it (would have) belonged, using digitized 1961 district maps. We then average outcomes by election-year
within those 1961 district boundaries.
Our primary outcome is the vote share received by Hindu nationalist parties in each district/election-
year. The case study literature identifies six main parties as having propounded a Hindu nationalist agenda
since independence: the Bharatiya Janata Party, the Bharatiya Jana Sangh, the Hindu Mahasabha, the
Ram Rajya Parishad, the Shiv Sena, and the 1977–1980 Janata coalition.15 Additional outcome measures,
and the steps taken to build them, are detailed in the results section and Online Appendix Table A2.
Online Appendix Table A1 gives summary statistics.
Estimation
Areas close to large pilgrimage sites might exhibit unusual demographic traits (e.g., be home to more very
religious people). Governments might also strategically time some religious gatherings to benefit from their
wide appeal. If true, naive OLS estimates based on either cross-sectional or longitudinal variation will be
biased. To overcome these inferential hurdles, we rely on a long time-series cross-sectional dataset, and the
plausibly exogenous timing of the Kumbh Mela festivals relative to India’s national elections.16 The core
15For the 1977 elections, which came in the aftermath of the Emergency launched by Prime Minister Indira
Gandhi, we code the Janata party coalition as Hindu nationalist. Hindu nationalist parties opted to stand under
the Janata banner that year, as part of a coordinated effort to unseat the Congress. Janata included a variety
of ideological tendencies. But it was repeatedly branded as Hindu nationalist in its fundamental orientation by
movement opponents (Jaffrelot 1999, 305). Internal disagreements over holding RSS membership were the primary
factor behind the coalition’s downfall. Strikingly, the coalition “gained electoral traction [at the 1977 Prayagraj
Kumbh Mela] because of unequivocal support of the Sadhu Samaj (the society of ascetics). The Dharma Sansad
(parliament of religion) and a vast majority of Sadhu Sammelans (saintly organizations) in the Kumbh of 1977 had
declared Indira as a mortal enemy of India” (Swarajya, February 28, 2013, bit.ly/41Q1y2u). We later confirm the
results’ robustness to recoding the main outcome measure for 1977 and 1980 based only on Janata candidates with
a Hindu nationalist background; see Supplementary Information C.
16See Online Appendix Oand Pfor further justification of this assumption. We identified no case of a national
election being timed with the Kumbh Mela in mind.
11
specification is a two-way fixed effects (TWFE) estimator of the form:
Ydt =β·Recent nearby Kumbhdt +γ0d+δt+θd(γ1d·t)+ϵdt (2)
where Yis the outcome. Time-invariant district attributes—such a region’s underlying political leanings,
institutional history, or legacies of Hindu nationalist organization—are captured by district fixed effects,
γd. Election fixed effects (δt) account for shocks such as party “waves” that are specific to each election
cycle. In some models we partial out district-specific linear time trends in the outcome variable: γd×t.
The idiosyncratic error term, ϵ, is clustered at the district level.
The parameter of interest is β. The key identification assumption is that, absent the Kumbh Mela,
districts near the events would have experienced the same trends in voting for Hindu nationalist parties
as districts further away. Crucially, generalized TWFE estimators are not equivalent to difference-in-
differences designs, and may suffer from two related problems (De Chaisemartin and d’Haultfoeuille 2020).
First, TWFE can place more weight on some treated cases rather than others as a function of treatment
timing. If treatment effects are heterogeneous and covary with treatment timing, the simple TWFE
procedure gives biased estimates of the average effect of treatment on the treated group. Second, the
weighted average effect may include negative weights resulting from “forbidden comparisons”: that is,
difference-in-differences comparisons of treated units against previously treated units, which are biased
if the effects of treatment are dynamic over time. Recognizing these concerns, we show that: (i) no
observations receive negative weights in our application; (ii) there are strong case-specific reasons to rule
out dynamic effects; and (iii) the results hold when using estimators that weight treated units equally and
exclude forbidden comparisons by design.
Results
Effects on Hindu nationalist vote share. The findings for the primary outcome are presented in
Table 1. Panel A, Column 1 reports that the occurrence of a nearby Kumbh Mela (<450kms by rail)
within one year prior to a Lok Sabha election increases the vote share for Hindu nationalist parties by 7.6
percentage points on average (p < 0.001). The inclusion of district-specific time trends in Panel A, Column
2 yields an effect estimate of 7.3 percentage points. These are sizable impacts relative to the overall outcome
mean of 21 percent. Implementing Conley (1999) adjustments for spatial correlation among districts leaves
standard errors substantively unchanged, even for radii as large as 1,500kms, which is half the maximum
rail distance between any two districts in the sample.
12
We perform an initial set of robustness checks. Panels B–D of Table 1display coefficient magnitudes
and statistical significance of a very similar order to the baseline model when (i) using the full (imbalanced)
panel data, (ii) restricting the estimation sample only to districts within 450kms of a Kumbh site (i.e., those
districts that could have been treated), and (iii) employing a continuous measure of Kumbh intensity.17
The effect is not driven by any one state or geographic region (Online Appendix Figure A5). Online
Appendix Table A3 shows that the results persist when coding the binary treatment using straight-line
distance to the Kumbh sites instead of rail distance, and when computing Hindu nationalist vote share for
1977 and 1980 using only Janata candidates with an avowedly Hindu nationalist background.
Regarding potential issues with the TWFE estimator, we find that there are no negative weights as-
sociated with the estimated ATTs (Table 1, Panel A).18 Results from the matrix completion method for
causal panel analysis—a class of counterfactual imputation estimators proposed by Athey et al. (2021) and
operationalized by Liu et al. (2022)—show an overall estimated effect of 6.7 percentage points (Table 1,
Panel F). The diagnostic placebo and parallel trends equivalence tests for this procedure—one, importantly,
that weights treated units equally and precludes forbidden comparisons—are also satisfied.19 Online Ap-
pendix Table A4 demonstrates the robustness of those findings for different codings of the binary treatment
variable.
Table 1, Panel E examines temporal heterogeneity. We interact the main treatment with an indicator
for the latter half of the time series, starting from 1984, which also marks the first Lok Sabha election
contested by the newly formed BJP. The estimated effect of a Recent nearby Kumbh is 8.9 percentage
17We compute a Kumbh “gravity” measure for each district/election year (with kindexing the four sites):
Kumbh time/rail distancedt =PK
i=1
Days elapsed since last Kumbhkt
Rail distance to Kumbh sitedt . Larger values within districts imply greater spa-
tial/temporal distance from—and thus less exposure to—the Kumbh Melas. We do not focus on this treatment
variable as generalized difference-in-differences estimators for continuous treatments involve complex assumptions.
18The absence of negative weights is due to our minimizing the possibility of dynamic effects by only considering
units to be “treated” when they occur within one year after a nearby Kumbh. Elections are, on average, spaced
more than four years apart. It seems unlikely that the effects of these events would still be changing five or more
years after they happen. Online Appendix Figure A4D supports this intuition: upon exiting treatment, previously
treated districts return to parallel trends, suggesting no dynamic effects.
19See Online Appendix Figure A4 for the default plots from the FEct R package used for this estimation. Note,
of the many estimators put forward for dealing with the issues surrounding TWFE, the Liu et al. (2022) framework
is most appropriate for our study because it permits treatment adoption to be both staggered and to switch on and
off.
13
points greater in the pre-1984 era than in the post-1984 era. This is unlikely to reflect a ceiling effect in
the later period: even in districts close to the Kumbh sites, the mean Hindu nationalist vote share has
not surpassed 34 percent (Online Appendix Figure A6). In Online Appendix Table A3, Panels D and
E, we show that the rapid growth of television and radio ownership after 1989 is not associated with an
attenuation of the Kumbh effect, as might be expected if media coverage of the events comes to wash
out the impacts of in-person attendance. We instead speculate that mass religious gatherings do more
to facilitate the expansion of religious party support earlier in the party’s life-cycle—when the resources
available for mobilization are scarcer, and the political opportunity structure is less welcoming.
Figure 2visualizes four last robustness tests. Panel Ai plots the effect of a Kumbh held within the past
year (compared to longer ago), across different distance bins, while Panel Aii plots the effect of a Kumbh
held up to 450kms away (compared to further away) across different time bins. For both regression models,
we observe the expected treatment-effect decay as—respectively—distance or time from the Kumbh grows.
Next, using the FEct matrix completion method, Panel B finds strong evidence of parallel pre-trends,
helping to validate the assumptions of the generalized difference-in-differences design. Each tile in Panel
C presents the main coefficient estimate from a regression using a differently specified binary treatment
variable—separately varying the time and distance cutoffs employed to code it. Reassuringly, we observe
a near-monotonic increase in the size of the estimated treatment effect as we shrink the time and distance
cutoffs set. Finally, in Panel D, we report placebo distributions generated from TWFE models using “fake”
versions of the treatment variable, variously permuting or randomly shuffling the timing of the Kumbh
Melas and/or their locations. We find that the result obtained from treatment based on the true Kumbh
schedule and location—indicated by the vertical dashed red lines—is highly unlikely to haven arisen by
chance. This test further mitigates concerns about spatial autocorrelation in the estimation of variance.
In short, the main finding of the paper is resilient. We now turn to theoretical mechanisms.
14
Table 1: This table reports estimates of the Kumbh Mela’s effects on Hindu nationalist vote share. The unit of
analysis is the 1961 district/election year. Data cover Lok Sabha elections from 1951–2019. Standard errors are
clustered by 1961 districts. Panels A–E present OLS regression estimates. Panel F presents estimates using the
FEct matrix completion method (which, note, does not enable the inclusion of linear time trends). Apart from Panel
D, all models employ the binary treatment Recent [<365 days] nearby [<450kms by rail] Kumbh. In Panel F, the
pre-trend test evaluates the null hypothesis that the 90 percent confidence intervals for the estimated ATTs in the
pretreatment periods, indexed by s, exceed 0.36 standard deviations of the residualized untreated outcome variable
(θ). The placebo test evaluates the null hypothesis that the 90 percent confidence intervals for estimated ATTs in a
placebo period p(here, two periods before treatment onset) exceed the same θ-defined equivalence range.
Outcome:
Hindu nationalist vote share (0-1)
(1) (2)
A. TWFE + binary treatment + balanced panel
Recent nearby Kumbh 0.076*** 0.073***
(0.009) (0.010)
N3,485 3,485
R20.68 0.77
1961 district FEs X X
Election year FEs X X
District-specific linear time trends X
Conley SE: 500 kms [0.0084] [0.0081]
Conley SE: 1,000 kms [0.0079] [0.0084]
Conley SE: 1,500 kms [0.0074] [0.0079]
Negative ATT weights 0%
B. TWFE + binary treatment + imbalanced panel
Recent nearby Kumbh 0.079*** 0.073***
(0.008) (0.008)
N4,881 4,881
R20.68 0.78
1961 district FEs X X
Election year FEs X X
District-specific linear time trends X
C. TWFE + binary treatment + balanced panel,
only districts w/in 450kms
Recent nearby Kumbh 0.084*** 0.063***
(0.013) (0.012)
N1,653 1,653
R20.67 0.77
1961 district FEs X X
Election year FEs X X
District-specific linear time trends X
D. TWFE + continuous treatment +
balanced panel
Ln. Kumbh time/rail distance -0.113*** -0.117***
(0.019) (0.020)
N3,485 3,485
R20.68 0.77
1961 district FEs X X
Election year FEs X X
District-specific linear time trends X
E. TWFE + binary treatment + balanced panel;
heterogeneous effects
Recent nearby Kumbh 0.109*** 0.100***
(0.013) (0.013)
Recent nearby Kumbh x Post-1984 -0.089*** -0.069***
(0.020) (0.017)
N3,485 3,485
R20.68 0.78
1961 district FEs X X
Election year FEs X X
District-specific linear time trends X
F. Counterfactual estimator, matrix completion method +
binary treatment + balanced panel
Recent nearby Kumbh 0.067***
(0.010)
Nunits 205
Pre-trend equivalence test (|ATTs|> θ, ∃s≤0), p-value: 0.000
Placebo equivalence test (|AT T p|> θ), p-value 0.000
*p<0.1; **p<0.05; ***p<0.01
15
(i) Estimated effect on Hindu
nationalist vote share of a Kumbh
occurring <1 year before an election
(ii) Estimated effect on Hindu
nationalist vote share of a Kumbh
occuring <450 kms away
0−450 kms
450−900 kms
900−1350 kms
1350−1800 kms
0−365 days
365−730 days
730−1095 days
1095−1460 days
0.00
0.05
0.10
Estimated effect
(percentage points)
A. Multiple binned treatment variables
(i) Pre−trends
−10 −8 −6 −4 −2 0
−0.10
−0.05
0.00
0.05
0.10
Election cycles relative to Kumbh
Estimated effect on Hindu nationalist
vote share (percentage points)
(ii) Post−ATT
Obs. Unit
Weighting
B. FEct matrix completion, main binary treatment
0.07 0.06 0.05 0.04
0.06 0.06 0.06 0.04
0.04 0.05 0.04 0.03
0.05 0.05 0.05 0.03
0.03 0.03 0.02 0.02
250
500
750
1000
400 800 1200 1600
Treatment distance cutoff
(Kumbh < X kms)
Treatment time cutoff
(Kumbh < Y days)
0.03
0.04
0.05
0.06
0.07
Estimated effect on
Hindu nationalist
vote share
(percentage points)
C. Varying the definition of binary treatment
−0.10 −0.05 0.00 0.05 0.10 −0.10 −0.05 0.00 0.05 0.10
0
20
40
60
80
0
50
100
0
1
2
3
0
30
60
90
Placebo treatment effect estimate (percentage points)
Frequency
D. Histograms of estimates based on placebo binary treatments
Figure 2: This figure presents tests of assumptions and robustness. Panel A: Results from two TWFE regressions estimating the effect of temporal and
spatial proximity to the Kumbh Mela on Hindu nationalist vote share, employing binned time and distance treatment indicators; estimating equations are
given in Online Appendix D.Panel B: Visualization of pre-trends and overall effect ATTs—for different weighting schemes—using the FEct matrix completion
method; see Online Appendix Section Ifor default plots. Panel C: Coefficient estimates from 20 TWFE regressions, each employing a differently coded version
of the treatment variable. Panel D: Distributions of estimated effects from multiple TWFE regressions, each using a different placebo treatment variable; the
vertical dashed line represents the estimated effect based on the true treatment variable; see Online Appendix Efor full description of the procedures. All
TWFE models in this figure include election year and 1961 district fixed effects. Confidence intervals are based on standard errors clustered by 1961 districts.
The “main” binary treatment is Recent [<365 days] nearby [<450kms by rail] Kumbh. All models employ the balanced-district panel of 3,485 observations.
16
Religiosity and identity change. We argued that mass religious events can help religious parties by
enlarging the base of religiously-inclined voters locally. To our knowledge, there are no publicly available,
large panel surveys for India that have asked directly about respondents’ religiosity. We rely instead on an
innovative, revealed preference measure of this concept.
A distinctive feature of Brahminical Hindu orthodoxy is lacto-vegetarianism: a diet that excludes
meat, fish, and eggs (Michaels 2004, 26). This, along with belief in karma,dharma, the Vedas, and a
pantheon of deities, is seen as constitutive of an “ideal” Hindu identity (Doniger 2010, 28). Importantly
for our purposes, vegetarianism is not mandated by India’s other major religions (Islam and Christianity).
Evaluating whether Kumbh Mela exposure popularizes strict vegetarian diets, therefore, can shed light on
the festival’s impact on a key private aspect of Hindu religiosity. Seen differently, it can illuminate whether
the Melas cause Hindus in general to adopt a largely upper-caste Hindu practice—a process Srinivas (1969)
calls “Sanskritization.” Promoting vegetarian diets has always been part of the Hindu nationalist political
agenda, too.20
Table 2: This table reports estimates of the Kumbh Mela’s effects on consumer expenditure. The unit of analysis
is the 1991 district/NSS round. Data cover all six “thick” NSS rounds that range from 1987–2012. Standard errors
are clustered by 1991 districts.
Outcome:
Prop. purchased meat,
fish, or eggs within
the last 30 days
(1) (2)
Recent [<365 days] nearby [<450kms] Kumbh -0.018** -0.024**
(0.009) (0.010)
N2,388 2,388
R20.92 0.95
1991 district FEs X X
NSS round FEs X X
District-specific linear time trends X
Negative ATT weights 0%
*p<0.1; **p<0.05; ***p<0.01
Leveraging the data from all quinquennial, “thick” rounds of surveys carried out by India’s National
Sample Survey Organization (NSSO), we construct a six-period, district-level dataset of household con-
sumption expenditures to test for Kumbh-induced changes in dietary practices. The surveys—the largest
of their kind—ask representative samples of households from nearly all districts in the country about their
spending on specific categories of food items over the past 30 days. We assign a household a value of
one if it reports having spent a non-zero amount of money on meat, fish, or eggs in that time period,
20“Vegetarianism: The Politics of Diet,” Frontline, July 23, 2018.
17
and zero otherwise. The final outcome measure is the average of that variable by district/survey round.
Using the same estimating approach laid out in Equation 2, Table 2, Columns 1 and 2 show that a Recent
nearby Kumbh reduces the proportion of district households reporting any expenditure on meat, fish, or
eggs by between 1.8 and 2.4 percentage points, buttressing our theory of religious identity change. Online
Appendix Table A5 attests to the result’s robustness, showing various specifications and placebo tests,
including that the effect materializes among Hindu households but not Muslim households, as we should
expect.
Platform co-optation and party organization. Our second mechanism stipulated that religious
parties enjoy an outsized advantage in using mass religious gatherings to recruit, organize, and campaign.
The 2001 Kumbh Mela in Prayagraj served as “an opportune backdrop for the Vishva Hindu Parishad
to organize its ninth meeting of spiritual leaders” and the event even helped “pole-vault [future prime
minister] Modi to national prominence” (Sitapati 2020, 258). At the 2019 Kumbh, “the BJP camp in the
mela ground had the biggest area and was no less than a temporary luxurious hotel.”21 We probe the
platform co-optation claim quantitatively.
To do so, we zero in on the consecutive Lok Sabha elections of 1967 and 1971. This pair of races
makes for a fortuitous case study. The 1971 election took place between March 1 and March 10, 1971,
immediately following the Ardh Kumbh in Prayagraj (which had ended in February). Party politics were
rife at the Mela grounds: “The Brahmacari’s camp was still going strong at the Ardh Kumbh Mela in
1971, ‘busy all the time blaring election propaganda against Indira Gandhi and her Congress’ ” (Maclean
2008, 213). There were no other Kumbhs besides the one at Prayagraj in the year prior to the 1971
election. Meanwhile, the last Kumbh before the 1967 election had occurred a full year previously. The
two elections each happen to have been accompanied by nationally representative, post-election opinion
polls, “the first serious empirical effort to study mass political behavior in India with national surveys”
(Eldersveld and Ahmed 1978, x). These coincidences allow for a two-wave difference-in-differences design.
Areas near to Prayagraj were “untreated” in 1967 and “treated” in 1971, while faraway areas remained
“untreated” in both election cycles. To understand the Kumbh’s impact on party organization, therefore,
we can simply assess whether Hindu nationalist party strength grew more between 1967 and 1971 in the
vicinity of Prayagraj than it did elsewhere.
The Bharatiya Jana Sangh (BJS) was the leading Hindu nationalist party at this time. Both the
21“Kumbh Country Turns into Battleground of Politics Ahead of LS Polls,” NewsClick, February 7, 2019,
bit.ly/2SMHzEj.
18
0.33***
0.25*** 0.19**
0.10
−0.03 −0.03 0.01 0.03 −0.02 0.00 −0.01 −0.05 −0.08
−0.17* −0.14 −0.14 −0.08 −0.06
−0.15 −0.13
−0.20 −0.19 −0.24* −0.22* −0.20
−0.06 −0.05
−0.12 −0.11
0.00 0.03 0.01 0.06 0.06
0.15 0.20
(B) Outcome: Congress organizational strength index
(A) Outcome: BJS organizational strength index
250 500 750 1000
−0.6
−0.3
0.0
0.3
0.6
−0.6
−0.3
0.0
0.3
0.6
Cutoff distance (kms) from Prayag used for coding Treatment indicator
Estimated ATT, 0−4 point scale
Figure 3: Each point estimate represents the coefficient on the interaction term from a separate TWFE regression,
following Equation 3. Models differ only in the treatment indicator, which is coded for different cutoff distances
shown on the horizontal axis. Confidence intervals are based on standard errors clustered by the pseudo-districts
described in Online Appendix M. There are 5,975 observations in each model. *p < 0.1; **p < 0.05; ***p < 0.01.
1967 and 1971 election surveys posed near-identical questions to respondents about their exposure to BJS
campaign activities in the lead-up to polling day. We infer the local strength of the BJS party organization
using four survey measures taken at the respondent level: whether they had received a BJS flier, whether
BJS canvassers had come to their doorstep, whether they had attended a BJS meeting, and whether they
were a BJS party member. We sum the binary variables together to make a four-point index of local Hindu
nationalist party strength. Then we run the following model multiple times:
Yidt =τ·(1{Prayagraj within cutoff distance by rail}d·1{1971 survey}t) + γd+δt+ϵidt (3)
Here, Yis the outcome reported by respondent iin spatial unit din election t, while γand δstand in for
unit and survey-round fixed effects, respectively.22
Figure 3A plots estimates of the coefficient of interest, τ, along with associated confidence intervals,
for models using different rail-distance cutoffs for coding the treatment indicator. We see clear evidence
that the BJS ground campaign and organization grew more in places close to Prayagraj between 1967
and 1971—to the tune of 0.19 to 0.33 scale points for areas within 300kms—over and above its growth
elsewhere. The mean of the outcome index was just 0.21 (out of 4) in the 1967 survey, suggesting that
22The process by which we linked sampling areas across the two survey rounds to make constant spatial units is
described in Online Appendix M.
19
very recent nearby exposure to the Kumbh more than doubled the party’s organizational capabilities.23
Importantly, Congress saw no such boost (Figure 3B), consistent with our claim that religious parties are
best positioned to cash in organizationally on mass religious events.
Polarization. What are the broader social and political implications of the Kumbh Mela? Social identity
theory maintains that self-esteem comes from belonging to social groups, and that a person’s self-esteem is
greater when their group is perceived as superior to other groups (Tajfel and Turner 1979). The entrench-
ment of Hindu self-identification caused by the Kumbhs might thus erode trust between religions, raising
the likelihood of communal conflict.24 Electoral polarization could also follow, as victimized groups adapt
their voting behavior to safeguard their interests (Rabushka and Shepsle 1972).
We first examine the Kumbh’s impact on the incidence and severity of intergroup violence. Using the
updated Varshney and Wilkinson (2006) dataset, we build a district/month panel of Hindu-Muslim riots
between 1951 and 2000 and generate the same Recent nearby Kumbh treatment indicator as before. Table
3uncovers no signs that the events escalate either the probability of any communal violence breaking out
(Columns 1–2) or the lethality of such violence (Columns 3–4).
Table 3: This table reports TWFE estimates of the Kumbh Mela’s effects on Hindu-Muslim riots. The unit of
analysis is the 1961 district/month. Data cover 1951–2000. Standard errors are clustered by 1961 districts.
Outcome:
Any riot Log(Num. killed + 1)
(1) (2) (3) (4)
Recent [<365 days] nearby [<450kms] Kumbh -0.001 -0.001 -0.001 -0.001
(0.001) (0.001) (0.001) (0.001)
N196,864 196,864 196,864 196,864
R20.04 0.04 0.03 0.04
1961 district FEs X X X X
Month FEs X X X X
District-specific linear time trends X X
Negative ATT weights 0% 0%
*p<0.1; **p<0.05; ***p<0.01
Next, single-member plurality voting—of the kind used in India—is thought to promote two-party com-
petition around the main axis of social division (Cox 1997). If the Melas make exclusionary religious parties
more successful, we might see a response from minority groups who risk losing out if Hindu nationalists win
office. Muslims are the population group most threatened by Hindu nationalist party rule. Consequently,
23The disaggregated results for the index component measures are shown in Online Appendix Figure A1.
24The organizational advances made by Hindu nationalists because of the Kumbhs could also contribute to “in-
stitutionalized riot systems” that stoke violence (Brass 2003).
20
(A) DV: Congress vote prop.
(B) DV: Hindu nationalist vote prop.
(C) DV: Other parties vote prop.
0 25 50 75 100 0 25 50 75 100 0 25 50 75 100
−0.2
−0.1
0.0
0.1
Share of district population Muslim, percentile ranked
Estimated marginal effect of
Recent nearby Kumbh
(percentage points)
Figure 4: This figure plots the estimated heterogeneous effects of Recent [<365 days] nearby [<450kms by rail]
Kumbh on three different outcomes according to the percentile-ranked share of Muslims in the district population,
as recorded in the 1961 Census of India. Data cover all Lok Sabha elections from 1951 to 2019. The unit of analysis
is the 1961 district/election year. Models include 1961 district and election-year fixed effects. Blue lines show
linear interactions, with 95 percent confidence bands. Red circles show estimated subgroup effects, with 95 percent
confidence intervals; subgroups are defined by low, middle, and high terciles of the moderator. Confidence intervals
are based on standard errors clustered by 1961 districts. All models employ the balanced-district panel of 3,485
observations.
we look at heterogeneity in the Kumbh treatment effect according to the fraction of the district population
that is Muslim. We focus separately on the vote share going to Hindu nationalist parties, the Congress
party—which has a reputation for being sensitive toward Muslim interests (Nellis et al. 2016, 259)—and
to all other parties.
Figure 4shows that Muslim population share is both positively and linearly associated with the mag-
nitude of the estimated Recent nearby Kumbh effect on Congress party support (Panel A). In districts
with the smallest shares of Muslims, Congress loses votes when Kumbh Melas are near in space and time.
But as the local Muslim population increases, the size of these losses decreases; indeed, in the top tercile
(according to percent Muslim), Congress even experiences small vote gains due to the Kumbh. There is no
comparable variability when Hindu nationalist party vote share is the outcome (Panel B). Panel C shows
that the Congress gains come at the expense of all other parties—that is, those whose brands are less
entwined with the religious/secular cleavage. The result is an ecological one; we cannot be certain that
individual Muslim voters are reacting defensively. Nevertheless, our best interpretation of Figure 4is that
Muslim communities, who fear marginalization under Hindu nationalist incumbency, consolidate around
the primary secularist party in anticipation of an electoral surge by Hindu nationalists.
21
Conclusion
This paper presents a new explanation for the electoral success of ethnoreligious parties. We show that
mass religious events, which are a cornerstone of religious practice worldwide, substantially boost the vote
shares of religiously-aligned parties. Employing a credible research design and focusing on India’s Kumbh
Mela, we find that a region’s spatial and temporal proximity to this pilgrimage festival increases support
for Hindu nationalist parties. Our theory posits—and our evidence suggests—that the effect operates via
changes in the religiosity of voters, and religious parties’ exploitation of the gatherings for organization-
building. Overall, the findings testify to the transformative consequences of large religious events for
politics and society.
Illuminating an overlooked factor behind the rise of majoritarianism in a country home to one sixth
of the global population is of direct interest. The theoretical framework we develop may apply to other
forms of collective religious worship, too, since church services, processions, and prayer groups are all
susceptible to platform co-optation by parties and politicians. An especially promising avenue for future
research would be to construct designs capable of decomposing the direct and indirect impacts of large
religious events—to gauge, say, how the effect of actual participation in them compares to their influence
when mediated through television or the internet. It would also be worth examining the degree to which
ritual gatherings in other non-Abrahamic religions, such as Buddhism and Shintoism, become politicized
as they do in contemporary Hinduism. Whether pluralist states should better regulate collective religious
observance to curb the spread of exclusionary political worldviews speaks to the much larger question of
religion’s appropriate place in secular democracy.
22
References
Agarwal, Ananay, Neelesh Agrawal, Saloni Bhogale, Sudheendra Hangal, Francesca Refsum
Jensenius, Mohit Kumar, Chinmay Narayan, Basim U Nissa, Priyamvada Trivedi, and
Gilles Verniers, “TCPD Indian Elections Data, Version 2.0,” Trivedi Centre for Political Data, Ashoka
University, 2021.
Athey, Susan, Mohsen Bayati, Nikolay Doudchenko, Guido Imbens, and Khashayar Khos-
ravi, “Matrix Completion Methods for Causal Panel Data Models,” Journal of the American Statistical
Association, 2021, 116 (536), 1716–1730.
Bapu, Prabhu,Hindu Mahasabha in Colonial North India, 1915-1930: Constructing Nation and History,
London: Routledge, 2013.
Bhela, Anita, “Globalization, Hinduism, and Cultural Change in India,” Asia Journal of Global Studies,
2010, 4(2), 93–102.
Brass, Paul,The Production of Hindu-Muslim Violence in Contemporary India, Seattle: University of
Washington Press, 2003.
Brooke, Steven,Winning Hearts and Votes: Social Services and the Islamist Political Advantage, Ithaca:
Cornell University Press, 2019.
Chaisemartin, Cl´ement De and Xavier d’Haultfoeuille, “Two-Way Fixed Effects Estimators with
Heterogeneous Treatment Effects,” American Economic Review, 2020, 110 (9), 2964–2996.
Chhibber, Pradeep and Sandeep Shastri,Religious Practice and Democracy in India, New York:
Cambridge University Press, 2014.
Christia, Fotini, Elizabeth Dekeyser, and Dean Knox, “The Nature and Origins of Sectarian
Animosity,” MIT Political Science Department Research Paper, 2019.
Chwe, Michael,Rational Ritual: Culture, Coordination, and Common Knowledge, Princeton: Princeton
University Press, 2013.
Clingingsmith, David, Asim Khwaja, and Michael Kremer, “Estimating the Impact of the Hajj:
Religion and Tolerance in Islam’s Global Gathering,” Quarterly Journal of Economics, 2009, 124 (3),
1133–1170.
Conley, Timothy, “GMM Estimation with Cross Sectional Dependence,” Journal of Econometrics, 1999,
92 (1), 1–45.
Cox, Gary,Making Votes Count: Strategic Coordination in the World’s Electoral Systems, New York:
Cambridge University Press, 1997.
Doellinger, David, “Prayers, Pilgrimages and Petitions: The Secret Church and the Growth of Civil
Society in Slovakia,” Nationalities Papers, 2002, 30 (2), 215–240.
Donaldson, Dave, “Railroads of the Raj: Estimating the Impact of Transportation Infrastructure,”
American Economic Review, 2018, 108 (4-5), 899–934.
Doniger, Wendy,The Hindus: An Alternative History, Oxford: Oxford University Press, 2010.
Durkheim, Emile,The Elementary Forms of the Religious Life, Oxford: Oxford University Press, 2008.
Eck, Diana,India: A Sacred Geography, New York: Three Rivers Press, 2012.
23
Eldersveld, Samuel and Bashiruddin Ahmed,Citizens and Politics: Mass Political Behavior in India,
Chicago: University of Chicago Press, 1978.
Esposito, John,The Islamic Threat: Myth or Reality?, Oxford: Oxford University Press, 1999.
Fox, Jonathan, “World Separation of Religion and State into the 21st Century,” Comparative Political
Studies, 2006, 39 (5), 537–569.
Fukuyama, Francis,Identity: The Demand for Dignity and the Politics of Resentment, New York:
Farrar, Straus and Giroux, 2018.
Gould, William,Hindu Nationalism and the Language of Politics in Late Colonial India, New York:
Cambridge University Press, 2004.
Grewal, Sharan, Amaney Jamal, Tarek Masoud, and Elizabeth Nugent, “Poverty and Divine
Rewards: The Electoral Advantage of Islamist Political Parties,” American Journal of Political Science,
2019, 63 (4), 859–874.
Jaffrelot, Christophe,The Hindu Nationalist Movement and Indian Politics, 1925 to the 1990s, New
Delhi: Penguin Books India, 1999.
Juergensmeyer, Mark,Violence and the Sacred in the Modern World, London: Routledge, 2019.
Kalla, Joshua and David Broockman, “The Minimal Persuasive Effects of Campaign Contact in
General Elections: Evidence from 49 Field Experiments,” American Political Science Review, 2018, 112
(1), 148–166.
Kalyvas, Stathis, “Commitment Problems in Emerging Democracies: The Case of Religious Parties,”
Comparative Politics, 2000, 32 (4), 379–398.
Khanna, Tarun, John Macomber, and Saloni Chaturvedi, “Kumbh Mela: India’s Pop-up Mega-
City,” Harvard Business School Case 214-023, 2013.
Lipset, Seymour and Stein Rokkan,Party Systems and Voter Alignments: Cross-National Perspec-
tives, New York: Free Press, 1967.
Liu, Licheng, Ye Wang, and Yiqing Xu, “A Practical Guide to Counterfactual Estimators for Causal
Inference with Time-Series Cross-Sectional Data,” American Journal of Political Science, 2022, advanced
online publication.
Lochtefeld, James, “The Construction of the Kumbha Mela,” South Asian Popular Culture, 2004, 2(2),
103–126.
Maclean, Kama,Pilgrimage and Power: The Kumbh Mela in Allahabad, 1765-1954, Oxford: Oxford
University Press, 2008.
McClendon, Gwyneth and Rachel Riedl,From Pews to Politics: Religious Sermons and Political
Participation in Africa, New York: Cambridge University Press, 2019.
Michaels, Axel,Hinduism: Past and Present, Princeton: Princeton University Press, 2004.
Nellis, Gareth, Michael Weaver, and Steven Rosenzweig, “Do Parties Matter for Ethnic Violence?
Evidence from India,” Quarterly Journal of Political Science, 2016, 11 (3), 249–277.
Nelson, Matthew,In the Shadow of Shari‘ah: Islam, Islamic Law, and Democracy in Pakistan, New
York: Columbia University Press, 2011.
24
Norris, Pippa and Ronald Inglehart,Sacred and Secular: Religion and Politics Worldwide, New York:
Cambridge University Press, 2011.
Olaveson, Tim, “Collective Effervescence and Communitas: Processual Models of Ritual and Society in
Emile Durkheim and Victor Turner,” Dialectical Anthropology, 2001, 26, 89–124.
Petrocik, John, “Issue Ownership in Presidential Elections, with a 1980 Case Study,” American Journal
of Political Science, 1996, 40 (3), 825–850.
Rabushka, Alvin and Kenneth Shepsle,Politics in Plural Societies: A Theory of Democratic Insta-
bility, Columbus: Charles E. Merrill, 1972.
Rudolph, Lloyd and Susanne Rudolph,The Modernity of Tradition: Political Development in India,
Chicago: University of Chicago Press, 1984.
Sherman, Taylor,Nehru’s India: A History in Seven Myths, Princeton: Princeton University Press,
2022.
Sitapati, Vinay,Jugalbandi: The BJP Before Modi, New Delhi: Penguin Random House India, 2020.
Srinivas, Mysore,Social Change in Modern India, Berkeley: University of California Press, 1969.
Tajfel, Henri and John Turner, “An Integrative Theory of Intergroup Conflict,” in W. Austin and
S. Worchel, eds., The Social Psychology of Inter-Group Relations, Monterey: Brooks/Cole, 1979, pp. 33–
47.
Tambiah, Stanley,Buddhism Betrayed?: