Populism à la Carte: Modi on Twitter

by J-T.M. · Published January 23, 2023 · Updated January 23, 2023

<< This preprint—SMUR version—is a manuscript published in the journal Global Policy. The SMUR version is archived on Academia.edu (incl. keywords). The citation can be downloaded from the Hal repository. This complies with the access policy of the journal. The original title of the article is: “Populism à la Carte: The paradoxical political communication of Narendra Modi on Twitter” (DOI:10.1111/1758-5899.13173). The article is co-authored with Vihang Jumle.

Table of contents
1.Abstract / 2.Introduction / 3.Populism and religion in the Global South / 4.Method and dataset / 5.Results and discussion / 6.Epilogue: Modi’s guruhood, an irenic populism / 7.References / 8.Appendix

Abstract

How and where is democracy “hacked”? Studies examining the variability of the populist discourse rely on the—often tacit—assumption that global digital platforms are used similarly from one country to the other. By examining political rhetoric on Twitter, we show that populist communication is tailored to a particular audience profile rather than a particular social media. More specifically, we compare Indian Prime Minister Narendra Modi’s tweets with his addresses on multiple mediums. We describe how Modi uses wide-reaching media such as radio to unfold a populist narrative based on institutional disintermediation, layman’s language, counselling and the staging of an intimate conversation with the masses. Modi uses more elitist and cosmopolitan social media sites in India, such as Twitter, to mitigate his populist credentials, and introduce himself as a respectable democratic leader favouring multilateral collaborations, banal nationalism and Gandhian peace-building. The variability of his political rhetoric across seven address formats indicates that Modi is populist tactically rather than ideologically, which complicates the tenets of the dominant ‘ideational’ populist paradigm. To further ground our argument empirically, we venture into comparing Modi and Trump Tweets. We suggest that due to the popularity of Twitter in the US, it was strategically useful for the former President of the United States to act populist on the platform, while that isn’t the case for Modi. This tends to indicate that not enough attention is given to inter-medium studies that complement comparisons between states/regions with comparisons nested within a coherent political space.

Keywords: populism, India, Twitter, Modi, political communication, representation

Introduction

The global rise of populism (Norris 2020) as well as the ubiquitous mediatisation of political communication (Mazzoleni 2008) have brought considerable scope for comparative studies on the use of language by elected leaders on social media, in particular the populists relying on unmediated (and anti-elite) appeals to the people. Research has shown how populists were using social media to personalise politics (Schroeder 2019) while circumventing hostile traditional media (Krämer 2018). Studies have highlighted the elective affinity between populism and social media. They emphasise how platforms’ attention economy favours eventually the para-institutional rhetoric of populists in which homophily—i.e. the bonding between similar people—is paramount (Engesser et al. 2017). “Rebellious” political styling (Gerbaudo 2018a) and permanent campaigning (Gerbaudo 2018b) against internal enemies (Kreiss and McGregor 2017) are two rhetorical traits of populists that work best on social media (SM). All in all, SM appear to be an opportunity structure—i.e. an empowering factor—for populists, because they facilitate “counter people” narratives (Sinha 2017) and help political figures to make political adverts more personal (Silva 2020).

Yet the current fixation with the global role of SM in helping populists win elections relies on two problematic assumptions. First, cross-country comparisons tend to neglect geographical variations in platform usage (Pal and Gonawela 2017). For example, in a country like the United States of America, 27.3 percent of the eligible population uses Twitter, while 2.2 percent of the Indian population—mostly educated, young, English-speaking and urban professionals—is on Twitter (GDR 2022a,b; Kumar 2019). Janus-faced Twitter on one side is a popular media, and on the other is an expert-first platform. As the populist narrative advocates for people-centrism, anti-elitism and restorative sovereignty (Ernst et al. 2017) by appealing specifically to the disenfranchised (Sears 2018), this prevailing blindness to actual audiences is questionable.

Moreover, our understanding of populist communication tends to assume homogeneity rather than specialisation. If a political figure is found populist on a social media site like Facebook, she is often deemed populist in toto. Kessel and Castelein (2016) for instance find that tweets reinforce the populist narrative set in speeches (see also Waisbord 2017). Ernst and colleagues (2019) argue that populist attitudes culminate in social media by exacerbating those found in TV talk shows. Populists would conceive online engagement as a “double-barreled” gun: Facebook being best suited to mobilise angry audiences, and Twitter as the most appropriate space to harass journalists (Jacobs and Sandberg 2020). To ensure a democratic digital future, we need to properly gauge trajectories of populist media communications. By underestimating the hybridity of any media system (Chadwick et al. 2015), we risk disregarding that populist attitudes of a given politician might not always co-occur across social media outlets (Ernst et al. 2017; Engesser et al. 2017).

This article addresses this lack of contextualisation of populist communication by examining the rhetoric of current Indian Prime Minister Narendra Modi (2014-) across various mediums. Seven channels of communication by Modi are considered: (1) tweets, (2) radio shows, (3) TV interviews, (4) general speeches, (5) commemoration addresses, (6) press statement and, (7) authored books. This approach is complemented by two types of comparisons: between Narendra Modi and former US president Donald Trump’s tweets as well as between Modi and other Indian political figures. We find that Modi’s language is characteristically populist in a range of media—radio in particular—but resolutely less on Twitter. As noted by others, Modi’s tweets appear to be banal, focusing on “partnerships and affiliations”, “ingratiation” and “feel-good, ritualised responses” (Pal 2015) as well as “postures of obeisance” (Rao 2018) towards foreign dignitaries, religious leaders, sportsmen, actors and party workers.

The article will identify the ceremonial footprint of Narendra Modi and discuss how he coalesces postures of thankfulness, celebration, wishes, felicitations, commemorations and collaborations with international counterparts. We will see how this abidance to an expected political etiquette sits uneasy with his overall populist image. Because political etiquette is normative (Sadovsky 2020), non-transgressive, and removed from popular praxis, it cannot be easily labelled as populist. This is reflective of Modi’s attitude on Twitter as a ‘normal’ executive head devoid of individual appeal of its own. This focus on political decorum is not resolutely populist, because populism necessitates the personalisation of political representation and the establishment of non-institutional rapports between leaders and led. We further suggest that since Modi does not need Twitter to bypass traditional media (which are in part controlled by his party), he engages with Twitter to target a distinctively cosmopolitan audience and present himself as a protocol-aware world leader actively engaging with his counterparts. In addition to distributing symbolic rewards to supporters by following them, Modi uses Twitter to point to his large spectrum of communication channels.

The article is divided into four sections. After (I) drawing on the contours of Modi’s populist politics and discussing existing literature on his political presence on Twitter, we (II) introduce the textual corpus and the method we used in the study. We subsequently (III) present our results and (IV) discuss their implications for the study of populism.

Modi, Populism and Twitter usage in India

Social media platforms are taking their hold on the Indian masses at a relatively slow pace. Although their influence in mass media and political discourse has grown, the effective reach of these networks among the population remains limited, with only a third of all citizens accessing social media (Kumar 2019). Furthermore, Twitter lags behind other social media outlets, with its growth tapering off around 20 percent after initial years of steady expansion.

As of 2019, Twitter continues to be the least popular of all major social networking sites with a usage (even sporadic) of 12 percent among Indian voters, while this number is above 30 percent for Facebook, WhatsApp and YouTube. Despite this, at a global level, nearly half of all scholarship on phenomena such as hate speech online has been on Twitter, sparking debates about the representativeness of the findings (Matamoros-Fernández 2021; Jungherr 2016). In the gladiatorial arena of social media, participation is not inclusive. In India, regular use of Twitter is upper caste, urban and male dominated. Upper castes are twice as likely to have high or moderate exposure to social media as Dalits and tribals. Twenty-four percent of women are likely to own a smartphone, compared with 41 percent for men (Kumar 2019). Five percent of those who studied up to matriculation are regular (every week) users of Twitter (contrary to 26 percent for Facebook and 33 percent for WhatsApp). Seventy-five percent of the population over 36 years old has no exposure to social media at all, and regular users of Twitter in this age group is around four percent (Ibid. 2019). Yet, half the population now has access to a smart phone, and India has among the world’s lowest data costs. Social media are thus a broken mirror, both highly reflective but also only a partial picture of a divided society (Mellon and Prosser 2017; Wojcik and Hughes 2019).

Despite this, Prime Minister (2014-) Narendra Modi has relied abundantly on Twitter to build a powerful brand online. Using Twitter as brand signalling, he crafted a tech-savvy, “high tech”, and pro-globalisation image of himself (Jaffrelot 2013; Kaur 2015; Ruparelia 2015; Sen 2016; Pal, Chandra and Vydiswaran 2016). In addition, Twitter has been employed by Modi to build political loyalty through following selective right-wing influencers (Manu 2020). Modi’s communication style on Twitter is characterised by the use of positive, uplifting messages, an avoidance of incendiary or direct conflict and the frequent use of wit and restraint (Pal et al. 2017). As noted by Rao (2018), Modi cultivates his image of a world leader on the medium, commenting on his interactions with heads of state, conveying birthday wishes to important personalities and increasing the frequency of his tweets when travelling abroad.

Modi’s tweets do not seem at first glance to be overtly populist. Because a populist figure projects a particular self-image as a commoner, her attitude in engaging with foreign affairs on Twitter should be confrontational and aimed at breaking with diplomatic conventions (Wojczewski 2020), which is not apparent in Modi’s tweets. This seems to contrast with authoritative accounts of Modi’s populist leadership, in which he displays exclusive personal strength, flatters popular Hindu identities and bypasses existing institutional and constitutional arrangements (Jaffrelot and Tillin 2017). Studies insist on the plebeian grammar displayed in Modi’s speeches and in his radio addresses, where he projects himself as a patriotic common man (Chakravartty and Roy 2015), a compassionate father (Srivastava 2015; Sinha 2017) and an intimate Hindu guru guiding one’s everyday life conduct [anonymised reference]. Considering this glaring discrepancy between the political communication of Narendra Modi on Twitter and his overall populist image, we turn to the empirical study of his discourse across a range of media.

Corpus and methods

Modi’s tweets were crawled using Twitter API product track for academic research, which allowed us to collect historical tweets. These amounted to 24,037 tweets between 01-02-2009 (Modi’s first tweet) and 16-06-2020 (date of data collection for Modi). While Modi predominantly tweeted in English, he often tweeted in Hindi, sometimes in other Indian regional languages and sparingly in foreign languages. Approximately 10 percent tweets were non-English and were therefore computationally translated to English and replaced within the dataset. On the other hand, Trump’s tweets were downloaded from the Trump Twitter Archive V2 repository since his account was no more accessible publicly on Twitter at the time of data collection (between 08-01-2021 and 20-11-2022). These amount to 56,572 tweets between 04-05-2009 and 08-01-2021. Alongside Modi’s tweets, we collected a wide range of Modi addresses (1,753 addresses) on the one hand, and a series of speeches and documents representative of Indian public life on the other (28 locutors; 355,216 addresses; 123,320,253 tokens). Together, they constitute an in-progress version of the DIPD—the Database of the Indian Political Discourse.

All data is preprocessed for analysis by ‘tokenising’ and ‘lemmatising’ it. Tokenising implies breaking individual tweets into small units (in this case individual words that form the tweets) making the text conducive for word-level analysis. Each tweet therefore is saved and studied as a collection of certain words, instead of a whole sentence. An individual word (the unit of analysis) is referred to as a ‘token’. As it is a standard practice in processing textual information, we lemmatise out tokenised words. Lemmatisation (shortly referred to as ‘lemma’) is the process of removing inflectional endings of different words and retaining only the base word (root word) or dictionary form of a word. For example ‘interact’ is the lemma of ‘interacted’, ‘interaction’, and so on. Stopwords are also removed whenever necessary, meaning we remove all those tokens (typically articles and pronouns) that are rather unimportant for textual analysis, retaining only the crux of a tweet. The process is akin to eliminating all low-level information token. The overview of all the figures presented here can be found in Appendix 0.

We are interested in describing the similarities and differences in language and vocabulary between Modi’s tweets and his other addresses by contrasting them against each other. Given that no one method is enough to fully capture the distinctiveness of Modi tweets, we use different computational methods to read the results in tandem. Since we are interested in studying Modi’s Twitter vocabulary, we first cluster the different words in his tweets. This gives us a sense of which words tend to appear together. A cluster analysis is an unsupervised method, meaning it requires no human-made labels, instead the method creates its own groups and assigns words to it based on their similarity (Hennig et al. 2016:4). We resort to descending hierarchical classification for our case (Bauer and Gaskell 2007:319).

Once clustered, we will be able to identify distinct classes with words in it. We then look at the address formats that contribute the most to such topics. Contribution can be understood as the relative importance of a given variable—locutors and address mediums—to the assignment of objects—words—to a cluster—classes of words in our case (Wierzchoń and Kłopotek 2018:9). The strength of the association between locutors-cum-medium and their respective class is measured using Pearson’s chi-square (χ2) test. Higher values indicate higher contributions.

We use topic modelling to analyse Modi and Trump sets of tweets. Methodologically, topic modelling takes all the tokens and assigns them a probability of occurring within each generated topic. We further create a dictionary of these tokens with the frequency of their occurrence in each document (i.e. in the tweet). We use a latent Dirichlet allocation approach (shortly ‘LDA’, Blei et al. 2003) to distribute Modi’s tokens into 17 topics and Trump’s tweets in 12 topics (with some topic overlaps in both cases). The selection of the number of topics was made on the coherence score of each batch of topics, with hyper parameters set to default. We are aware that there are arguments on either side that support and criticise the performance of LDA on short texts, sometimes making the method’s efficacy case specific (Jonsson and Stolee 2016; Negara et al. 2019; Yan et al. 2013). We therefore remain cognizant of reading the results as indicative and in light of results from other methods and not qualitatively over-interpreting the topics.

While comparing Modi’s topics to that of Trump topics on Twitter gives us a good sense of the different ‘personalities’ they intend to create for their medium specific audiences, comparing within Modi’s different addresses is also a confirmatory way to show that the changes in his vocabulary are based on the chosen medium. To demonstrate this, we compute specificity scores for all the tokens used by him in different addresses. Weighted frequencies and specificity scores have some similarities, but the latter is more accurate in defining the significance of a certain token frequency in sub-part of a textual corpus (Pincemin 2012, Gréa 2017). In our case, these sub-parts are formed by Modi’s different addresses. The higher the score, the higher the use of a certain token in a particular address. To facilitate the reading of graphs, scores are capped between 300 and -300 (across mediums), making each token’s relative significance comparable across addresses.

We are particularly interested in studying the overall similarities between Modi’s tweets and his other forms of addresses. We are also comparing Modi’s interventions with oratory renditions and textual outputs (which we term documents, see Lau and Baldwin 2016) produced by a range of other Indian political locutors. To this end, we employed the Doc2Vec algorithm (Le and Mikolov 2014) to compute cosine similarities (Sharaki 2020). The numeric vectors generated by the Doc2Vec algorithm provide a sense of different token’s neighbouring tokens and their universal context, allowing us to infer similarities between different kinds of documents, often termed cosine similarity. What makes the Doc2Vec stand out is its ability to look at words in context to compute document similarities. The output is a number between 0 and 1. Each form of document is assigned a number with respect to another form of document.

Before turning to the analysis per se, we use in Figure 1 below Correspondence Analysis (CA) to have a preliminary visualisation of the distribution of lemmas used by Modi across his address formats. CA is similar to cross-tabulations; it is based on a frequency distribution of two variables: in our case Modi’s address formats on the one hand, and Modi’s lemmas on the other. CA however gives us the possibility not only to measure, but to visualize ‘distances’ (degrees of difference) between the entries of these two variables (Blasius and Greenacre 2006). We use this method to compute distances from expected word frequencies for each address format in the Modi corpus. This gives us a first sense of the relationship between these units—i.e. how similar his tweets are from the rest. Lemmas are in blue, and address formats in red. Figure 1 (top graph) indicates that tweets are outliers when compared with other address formats. The bottom graph gives a first sense of the lemmas which are more specific to Modi tweets.

Analysis

We are here interested in language variation. The first step of the analysis builds on the major themes found in the Modi corpus. As indicated in the methods section, we use descending hierarchical classification to group lexicon frequently occurring together. Figure 2 (right side) displays the main classes retrieved by the algorithm. We then estimate the chi-square (x2) contribution of each address format to these classes. Results are displayed in on the left side of the figure. The heatmap provides insights on the classes most contributed by Modi Tweets: the 5th and 6th one. In Class 5, we retrieved lemmas like birthday, tribute, pray, congratulation, president, anniversary, condolence, jayanti. In Class 6 we identified tokens such as president, thank, meet, share, visit, relation, summit, welcome, bilateral, interact, delight. Class 5 is dominated by the theme of commemorations and state honours, while Class 6 emphasises various cooperations with international and national stakeholders. Both classes are imbued with protocol-related terms and are weaved around the formal and ritualistic function of Indian Prime Ministers (Bajpai 2018).

Due to length limitations, we restricted the analysis to the study of Modi’s rhetoric on Twitter rather than focusing on his holistic use of the platform. For instance, we did not examine how he follows celebrities, right leaning influencers, trolls and online hate-spreaders as a strategy to implicitly reward their activity. We believe these aspects have been covered extensively by Pal and colleagues (2016), Sinha (2017), Rao (2018), Akbar and colleagues (2020), as well as Bhat and Chadha (2022). As shown by Rodrigues and Niemann (2017), despite his use of Twitter as a one-way conversation, Modi managed to associate his flagship “clean India” campaign to well-known media influencers—including film stars and sportsmen. The analysis also does not concern itself with occasional incidents like alleged hacks of his account and the tweets made by impersonators. It could be argued that such incidents were not involuntary and exogenous but part of a systematic communication strategy. However, there is no certainty about this, and we could verify these claims. Retrospectively such tweets have been retracted with clarificatory notes. The analysis therefore does not consider these occasional and numerically insignificant (a couple tweets in a pool of thousands) tweets as part of Modi’s broad Twitter communication strategy.

This exploratory analysis indicates that Modi might be using tweets to distribute symbolic rewards, invoke historical figures and portray himself as a legitimate executive head working collaboratively at a global stage. Modi’s tweets do not directly highlight his distinctiveness as a politician and policy maker. They instead rely on affability and compliance with both the international order and national customs. Consequently, Class 3, which captures Modi’s personal and ‘intimate’ lexical field is significantly underused in his tweets. This Class is indeed dominated by perceptual vocabulary (feel , think ), a lexicon of parenthood (child, kid, parent, interact ) and lemmas related to everyday life (habit, doctor ). The latter register is indicative of a classic populist trope, where commonness and homophily are invoked to flaunt a disintermediated relationship between leaders and led. By precluding this vocabulary, his tweets reify a non-personalistic Prime Ministerial function, ultimately underplaying his populist ethos on other mediums. We further examine this assumption through comparing dominant themes in Modi’s tweets with those of a populist par excellence: Donald Trump.

In Figure 3, we compare the four most prominent topics in both Modi (on the left) and Trump’s tweets (on the right). We colour-coded the vocabulary belonging to each topic. The full list of lemmas in each topic can be found in Appendix 1. On the horizontal axis, the frequency for each lemma is indicated, and on the vertical axis, we report how specific to their topics lemmas are. A value of 1 means that 100 percent of the occurrences of a particular lemma have been classified in a particular topic.

Both politicians display a topic—the red one—that contains people-centric vocabulary. These two topics include the lemmas people and corrupt. Despite this superficial similarity, Trump’s tweets are the only ones clearly identifying an enemy of the people, in this case mass media in the United States (e.g., fake, media, witch, hunt). Modi’s tweets do not explicitly display the typical Manichean views spread by populists (Mudde 2004). The lexicon of Modi’s blue topic displays inclusive and vague terms such as nation, India, women, team, bharat but does not engage with those putative threats to the people characteristic of populist speeches. For instance, by not explicitly engaging with anti-Muslim tropes despite the avowed communal politics of his party, he marks his difference from Trump who was explicit about another form of insiders-outsiders discrimination—his anti-immigration agenda (wall, immigration, border, crime). Explicit black and white agonisitic tropes (Mouffe 2013) are central to populist narratives, yet they tend to be implicit if not absent in most tweets by Narendra Modi. To better identify this pattern, we turn to a more detailed analysis of those words which are most represented (and underrepresented) in Modi’s tweets as compared to his other address formats.

Figure 4 visualises those words which are most specific to Modi tweets as compared to his other address formats. We display those words which have the highest (left side of the graph) and lowest (right) specific scores in tweets—for display purposes, we capped scores to 300. Scores for tweets are shown with triangle-shaped plotting symbols. This offers scope for several observations. First and foremost, the ceremonial and multilateral-prone language epitomising Modi’s tweets (wishes, ties, Sharing, Congratulations, meeting, greetings) is rarely found in his speeches (blue dots). Examining the right side of the plot is even more informative. Importantly, the vocabulary used by Modi to address Indian constituents directly (country, Friends, sisters, village, house, people) is not often present in his tweets.

Additionally, it appears that welfare measures—and by extension, internal political matters—tend to be evoked less often in his tweets than in his speeches (and Independence Day addresses). Words such as electricity, scheme, money, bank, gas, water point more specifically at the developmental and redistribution-driven points of the Modi government. The relative absence of this vocabulary from his tweets tends to indicate that he does not use the platform to convey messages to the Indian masses. As a result, the repertoire traditionally associated with right-wing electoral politics (lotus, Janata, sisters, Janata, terrorists, Sangh, land, Guruji, army) is found more prominently in other mediums than Twitter. This is indicative that Modi considers the micro-blogging platform as a non-electoral tool to conduct Foreign Policy, build soft power and bolster his reputation as a transnational leader. Lastly, abundant mentions of other media in his tweets (mostly his radio show Mann ki baats and his Facebook posts) indicate that Modi uses Twitter as a hub to advertise his content hosted on other platforms.

From the above account, it emerges that Modi does not use Twitter to enforce his populist narrative, which is better exemplified in his radio addresses and public speeches. To further inquire about this assertion, we examine in Figure 5 how similar the seven Modi addresses are to a range of other political, administrative, and religious figures of the Indian public landscape since the end of the nineteenth century. To do so, we use the algorithm Doc2Vec described in the methodology section. On the right side of the radar chart we have gathered addresses that that are reputed to be technocratic in nature. This includes a variety of administrative documents, answers of government ministers in Parliament (1999-2019), speeches by the governor of the Reserve Bank of India (1990-20), public addresses by economist-turned-Prime Minister (Baru 2014) Manmohan Singh (2004-2014), and select administrative releases (e.g., Five Year Plans, 1951-12). The fact that much of this oratory and textual material (in particular ministerial answers in Parliament and Manmohan Singh speeches) are ‘more similar’ to Modi’s Tweets than his other addresses indicates that his tweets have more technocratic than populist attributes. Contrary to the acrimonious ‘experts-vs-people’ narrative of populists, “expertocratic” language aims at neutralising political conflict and rule in the name of expertise rather than popular will (Caramani 2017).

As evident in Figure 6, Modi uses Twitter to claim the legacy of several historical figures traditionally opposed to the Hindu right he belongs to (Phadnis and Kashyap 2019). By doing so, Modi employs Twitter to appear as a non-controversial figure. We have represented below the distribution (i.e. using specificity scores) of the mentions of 20 of the most important Indian historical personalities in Modi’s various address formats. While figures such as gurus, swamis or babas are neither overrepresented nor underrepresented in his tweets as compared to other Modi’s speech formats, Gandhi—a figure abhorred by Hindu nationalists—is most frequently invoked in his tweets (1,153 mentions, score=29.1). To be sure, Gandhi in Modi’s mouth is not invoked from an historical perspective by chronicling his fights for minority rights or his rejection of nationalist desires based on confessional identities. Instead—as shown by the co-occurrents of the word Gandhi in his tweets, he is merely associated to official celebrations, and symbols of nationhood, as the use of words 150th (score=8), pay (8), anniversary (7), devotional (6), monument (4), Father (4), tribute (3) indicate. Interestingly, Modi’s use of Gandhi in his Mann ki baats (radio shows) is much more associated to vague, patriotic, and ahistorical political values. There, he is re-purposed as a generic symbol of freedom (7), cleanliness (sanitation, 7; cleanliness, 2) and dream (3). In his Mann ki baat programmes, the frequent association of the word Gandhi with the word mantra is indicative of how Gandhi is used in an incantatory manner, as a mascot for Modi’s Swachh Bharat (clean India) campaign on sanitation, as “India’s biggest brand ambassador” (BS 2013) for external relations and as a slogan for positive values such as peace and harmony.

Discussion and conclusion

As we have outlined, the impression management strategy of Narendra Modi on Twitter is characterised by ‘ingratiation’. We have displayed a few tweets exemplifying this stylistic trope in Appendix 2. In this case, the logic of ingratiation is marked by a complimentary rhetoric of other-enhancement, flattery, modesty and honour-giving in which the Prime-Ministerial function rather than his persona is emphasised (Jackson and Lilleker 2011). Ingratiation is used to generate favourable and feel-good responses that circumvent direct conflicts and appear to be ill-suited to any Manichean populist narrative. We suggest that the barely-populist language of Modi on Twitter is best understood when placing his tweets side by side with his other addresses.

Contrary to established knowledge on populism and social media, the populist communication of Narendra Modi does not use the microblogging platform to directly address the plebes. As indicated in Figure 7, Modi uses markers of populist intimate connect with constituents during his radio shows Mann ki baats (MKBs) rather than in his tweets. Among the words that are both overrepresented in MKBs (>10) and underrepresented in his Tweets (<10), we find I, feel, me, you, countrymen, sense. Contrary to his radio shows, Modi does not use tweets to individualise his personal image, invoke feelings of proximity and theatrise unmediated connect with the common people. In line with our analysis, we can claim that Modi is much more populist on radio, where his messages travel far and wide, than on Twitter.

This result can be better understood when considering the Twitter audience in India, which—as we have seen—can be regarded as ‘elitist’. Because of this, the public on the platform might be only moderately receptive to the anti-establishment discourse proposed by populists. As noted by Bajpai (2021), radio in India is a much more suitable outreach channel to appeal to popular sections. Modi himself, in his first Mann ki baat stressed on this aspect: “I am glad to talk with you through this simple medium of Radio, which serves every corner of the nation. I can reach the poorest homes, as mine, my nation’s strength lies with the mothers, sisters and youth; my nation’s strength lies with the farmers” (Modi 2019:234). The viewership of Mann ki baats would be approximately 40 million listeners per episode (ET 2017), mainly because the show is relayed by a range of public (34 in total) and private (91) channels (The Wire 2021). The exceptional reach of the show beyond and across digital divides makes it a formidable tool for the deployment of a populist narrative. The radio listeners, through experiencing closeness with the speaker, feel like they are drinking from a fire hose.

Five important insights can be derived from this analysis. First, populism might not always be considered as an idea, but also as a strategy that can be selectively implemented by political leaders, depending on the actual audience they target. The variability of the populist language mobilised by Narendra Modi across media is an excellent example of the professionalisation of political communication in the Global South. In this case scenario, Modi’s political discourse is tailored to different audience segments for which he adapts both the content and the style of the message he addresses to them. By adopting substantially different language registers on Twitter than on his other speeches, Modi demonstrates how specialised and audience-specific his political communication is.

Second, social media emerges from this account as versatile spaces which do not necessarily serve the function of populist vehicles. As we have seen, some populists prefer to perform as such on non-digital mediums rather than on social media, which they use to legitimate themselves as respectable political leaders. Consequently, the elective affinity between populists and social media cannot be taken as granted. Contra structuralist and functionalist understandings of this interplay, we suggest that the feedback loop between populist communication and social media is at best conditional, and the necessary condition is that a given platform should not be used to target niche audiences and specialised public. When a social media can break the hold of traditional media and access directly a popular audience, they often become useful to populists as a pitch for political anger. However, when a populist in power already controls mainstream media, and when social media are monopolised by educated professionals, social media tend to disincentivise the use of populist language.

Third, scholars of social media often assume that platforms are ‘practiced’ homogenously across regional contexts. As the assumption goes, Twitter consumption in—for example—the US and Twitter in India is sufficiently similar and ubiquitous to enable cross-country comparisons of the discourses they harbour. We have seen that this assumption could be misleading, as it flattens the often-abysmal demographic and sociological variations in media use across territories. By disregarding local contexts, comparativists run the risk of comparing what is incomparable. In our case the problem was to contrast populist discourses on two different social media platforms: a popular and a specialised one. If we were to compare Modi and Trump on Twitter, we would find that only the former US president is populist. However, regional experts know that both politicians are populists, as they try to appeal directly to common masses by emulating closeness, simplicity, and the rejection of established elites. They both want the be ‘like the people’, but Modi prefers to use radio over social media to unfold his populist communication. This finding calls for comparing differently, by inviting more studies on how the populist rhetoric varies substantially from one media to the other according to the social composition of target audiences.

Fourth, Modi’s lack of populist content on Twitter does not necessarily imply that his use of Twitter is devoid of any populism. For instance, Modi’s foreign policy projects his closeness to world leaders on the platform, suggesting he has an intimate and unceremonial—hence populist—reach to the representatives of powerful nations. Such mise-en-scène is best represented by his habit to frequently hug his counterparts, especially if they are from the Global North (Chakrabarti et al. 2019), although we did not find evidence that Modi favours Twitter as the showcasing platform for his ‘populist hugs’. To be clear, communication around Modi’s social media tech-savviness can be seen as populist, while his social media content is not necessarily so. We suggest that this is not contradictory, as narratives of social-media use can be propagated on popular media (Jaffrelot 2013), while confining the non-populist content to the platform itself.

Fifth, we would like to acknowledge that the binary populism/non-populism in Modi’s discourse is not always structured around media lines. Twitter’s inherent intertextuality (via embedded links, hashtags etc.) (Davis 2013) allows Modi to redirect users to his populist content hosted outside of the platform. Similarly, excerpts and adaptations of his Mann ki baats can be found on Twitter, on television and on his website. Relatedly, while statesman-like content on Twitter is not per se populist, it can support other forms of populist rhetoric. For instance, Modi’s tropes fostering his international acceptability are congruent with his populist speeches that instil feelings of vishwas (trust) through postures of commonness and cultural relatedness (Sircar 2020). While non-populist, Modi’s content on Twitter helps him assert more populist narratives on other media in ways that are not always easy to capture in a statistical fashion due to the interdiscursive character of his addresses. Not only Modi speeches acquire different meanings according to a given political context, but they can signify different things for different audiences. This is particularly true for his Mann ki baat, whose positivity can be read as negative indictments against the minority community, especially when they are relayed by BJP politicians such as Amit Shah, Pragya Thakur and Yogi Adityanath. While Modi’s numerous references to Gandhi usually evoke an a-historical symbol of cleanliness, they also contain subliminal threats to the minority community via the imagery of cleansing.

References

Akbar, S. Z. et al. (2020) Anatomy of a Rumour: Social media and the suicide of Sushant Singh Rajput. arXiv (preprint). [Online]. Available from: https://arxiv.org/abs/2009.11744. (Accessed 12 December 2022).

Bajpai, A. (2021) ‘“Matters of the Heart”: The Sentimental Indian Prime Minister on All India Radio’, in Christoph Kohl et al. (eds.) The politics of authenticity and populist discourses: media and education in Brazil, India and Ukraine. Global political sociology. Cham, Switzerland: Palgrave Macmillan. pp. 140–167.

Bajpai, A. (2018) Speaking the nation: the oratorical making of secular, neoliberal India. First edition. New Delhi, India: Oxford University Press.

Baru, S. (2014) The accidental prime minister: the making and unmaking of Manmohan Singh. New Delhi: Penguin.

Bauer, M. & Gaskell, G. (eds.) (2007) Qualitative Researching with Text, Image and Sound: A Practical Handbook for Social Research. London; Thousand Oaks, Calif: SAGE.

Bhat, P., & Chadha, K. (2022). The mob, the state and harassment of journalists via twitter in India. Digital Journalism, 1–21. [Online]. Available from: https://doi.org/10.1080/21670811.2022.2134164 (Accessed 12 December 2022).

BS: Business Standard (2013) Modi shares his ‘Brand India’ mantra with advertising executives. Business Standard India. 30 September. [online]. Available from: https://www.business-standard.com/article/news-ani/modi-shares-his-brand-india-mantra-with-advertising-executives-113093001093_1.html (Accessed 2 April 2022).

Caramani, D. (2017) Will vs. Reason: The Populist and Technocratic Forms of Political Representation and Their Critique to Party Government. American Political Science Review. 111 (1), 54–67.

Chadwick, A. et al. (2015) ‘Politics in the Age of Hybrid Media’, in Axel Bruns et al. (eds.) The Routledge Companion to Social Media and Politics. 1st edition. New York, NY: Routledge, 2016.: Routledge. pp. 7–22.

Chakravartty, P. & Roy, S. (2015) Mr. Modi Goes to Delhi: Mediated Populism and the 2014 Indian Elections. Television & New Media. 16 (4), 311–322.

Engesser, S., Ernst, N., et al. (2017) Populism and social media: how politicians spread a fragmented ideology. Information, Communication & Society. 20 (8), 1109–1126.

Engesser, S., Fawzi, N., et al. (2017) Populist online communication: introduction to the special issue. Information, Communication & Society. 20 (9), 1279–1292.

Ernst, N. et al. (2019) Populists Prefer Social Media Over Talk Shows: An Analysis of Populist Messages and Stylistic Elements Across Six Countries. Social Media + Society. 5 (1), 205-233.

GDR: Global Digital Reports (2022a) Digital 2022: India. [online]. Available from: https://datareportal.com/reports/digital-2022-india (Accessed 2 April 2022).

GDR: Global Digital Reports (2022b) Digital 2022: The United States of America. [online]. Available from: https://datareportal.com/reports/digital-2022-united-states-of-america (Accessed 2 April 2022).

Gerbaudo, P. (2018) Social media and populism: an elective affinity? Media, Culture & Society. 40 (5), 745–753.

Gerbaudo, P. (2019) The digital party: political organisation and online democracy. Digital barricades: interventions in digital culture and politics. London: Pluto Press.

Gréa, P. (2017) Inside in French. Cognitive Linguistics. 28 (1), 77–130.

Greenacre, M. J. & Blasius, J. (2006) ‘Correspondence Analysis and Related Methods in Practice’, in Michael J. Greenacre & Jörg Blasius (eds.) Multiple correspondence analysis and related methods. Statistics in the social and behavioral sciences series. Boca Raton: Chapman & Hall/CRC. pp. 3–40.

Hennig, C. M. et al. (eds.) (2016) Handbook of cluster analysis. Chapman & Hall/CRC handbooks of modern statistical methods 9. Boca Raton: CRC Press, Taylor & Francis Group.

Jacobs, K. et al. (2020) Twitter and Facebook: Populists’ double-barreled gun? New Media & Society. 22 (4), 611–633.

Jaffrelot, C. (2013) Gujarat Elections: The Sub-Text of Modi’s ‘Hattrick’—High Tech Populism and the ‘Neo-middle Class’. Studies in Indian Politics. 1 (1), 79–95.

Jaffrelot, C. & Tillin, L. (2017) ‘Populism in India’, in Cristóbal Rovira Kaltwasser et al. (eds.) The Oxford Handbook of Populism. Oxford University Press. pp. 321–347.

Jónsson, E. & Stolee, J. (2016) An Evaluation of Topic Modelling Techniques for Twitter. CS Toronto preprints. 1–11.

Jungherr, A. (2016) Twitter use in election campaigns: A systematic literature review. Journal of Information Technology & Politics. 13 (1), 72–91.

Kaur, R. (2015) Good Times, Brought to You by Brand Modi. Television & New Media. 16 (4), 323–330.

Kessel, S. van & Castelein, R. (2016) Shifting the blame. Populist politicians’ use of Twitter as a tool of opposition. Journal of Contemporary European Research. 12 (2), 1–21.

Krämer, B. (2018) Populism, Media, and the Form of Society. Communication Theory. 28 (4), 444–465.

Kreiss, D. & Mcgregor, S. C. (2018) Technology Firms Shape Political Communication: The Work of Microsoft, Facebook, Twitter, and Google With Campaigns During the 2016 U.S. Presidential Cycle. Political Communication. 35 (2), 155–177.

Kumar, S. (2019) Social Media and Political Behaviour. [online]. Available from: https://rb.gy/emedy8 (Accessed 2 April 2022).

Lau, J. H. & Baldwin, T. (2016) ‘An Empirical Evaluation of doc2vec with Practical Insights into Document Embedding Generation’, in Proceedings of the 1st Workshop on Representation Learning for NLP. 2016 Berlin, Germany: Association for Computational Linguistics. pp. 78–86. [online]. Available from: http://aclweb.org/anthology/W16-1609 (Accessed 2 April 2022).

Le, Q. & Mikolov, T. (2014) ‘Distributed representations of sentences and documents’, in Proceedings of the 31st International Conference on International Conference on Machine Learning – Volume 32. ICML’14. 21 June 2014 Beijing, China: JMLR.org. p. II-1188-II–1196.

Manu, D. (2020) Lok Sabha Elections 2019: Analysing the Online Political Battlefield in India. Master of Science thesis. Hyderabad: International Institute of Information Technology Hyderabad.

Matamoros-Fernández, A. & Farkas, J. (2021) Racism, Hate Speech, and Social Media: A Systematic Review and Critique. Television & New Media. 22 (2), 205–224.

Mazzoleni, G. (2008) ‘Populism and the Media’, in Daniele Albertazzi & Duncan McDonnell (eds.) Twenty-First Century Populism. London: Palgrave Macmillan UK. pp. 49–64.

Mellon, J. & Prosser, C. (2017) Twitter and Facebook are not representative of the general population: Political attitudes and demographics of British social media users. Research & Politics. 4 (3).

Modī, N. (2019) Mann ki baat: a social revolution on radio. First impression. New Delhi: Rupa.

Mouffe, C. et al. (2013) Agonistics: thinking the world politically. London; New York: Verso.

Negara, E. S. et al. (2019) ‘Topic Modelling Twitter Data with Latent Dirichlet Allocation Method’, in 2019 International Conference on Electrical Engineering and Computer Science (ICECOS). October 2019 Batam Island, Indonesia: IEEE. pp. 386–390. [online]. Available from: https://ieeexplore.ieee.org/document/8984523/ (Accessed 2 April 2022).

Norris, P. (2020) Measuring populism worldwide. Party Politics. 26 (6), 697–717.

Pal, J. (2015) Banalities Turned Viral: Narendra Modi and the Political Tweet. Television & New Media. 16 (4), 378–387.

Pal, J. et al. (2016) Twitter and the Rebranding of Narendra Modi. Economic and Political Weekly. 51 (8), 52-60.

Pal, J. et al. (2017) Mediatized Populisms| Innuendo as Outreach: @narendramodi and the Use of Political Irony on Twitter. International Journal of Communication. 11 (0), 22.

Pal, J. et al. (2015) Twitter and the Rebranding of Narendra Modi. Economic and Political Weekly. 51 (8), 7–8.

Pal, J. & Gonawela, A. (2017) Studying political communication on Twitter: the case for small data. Current Opinion in Behavioral Sciences. 1897–102.

Phadnis, A. & Kashyap, A. (2019) The Politics of Historical Personalities: Textual Analysis of Speeches by the Indian Prime Minister.

Pincemin, B. (2012) Hétérogénéité des corpus et textométrie. Langages. n° 187 (3), 13–26.

Rao, S. (2018) Making of Selfie Nationalism: Narendra Modi, the Paradigm Shift to Social Media Governance, and Crisis of Democracy. Journal of Communication Inquiry. 42 (2), 166–183.

Rodrigues, U., & Niemann, M. (2017). Social Media as a Platform for Incessant Political Communication: A Case Study of Modi’s “Clean India” Campaign. International Journal of Communication. 11 (23).

Ruparelia, S. (2015) ‘Minimum Government, Maximum Governance’: The Restructuring of Power in Modi’s India. South Asia: Journal of South Asian Studies. 38 (4), 755–775.

Schroeder, R. (2019) Digital Media and the Entrenchment of Right-Wing Populist Agendas. Social Media + Society. 5 (4).

Sen, R. (2016) Narendra Modi’s makeover and the politics of symbolism. Journal of Asian Public Policy. 9 (2), 98–111.

Sharaki, O. (2020) Detecting Document Similarity With Doc2vec [online]. Available from: https://towardsdatascience.com/detecting-document-similarity-with-doc2vec-f8289a9a7db7 (Accessed 2 April 2022).

Silva, M. et al. (2020) Facebook Ads Monitor: An Independent Auditing System for Political Ads on Facebook. [online]. Available from: https://arxiv.org/abs/2001.10581 (Accessed 1 April 2022).

Sinha, S. (2017) Mediatized Populisms| Fragile Hegemony: Social Media and Competitive Electoral Populism in India. International Journal of Communication. 11 (0), 23.

Srivastava, S. (2015) Modi-Masculinity: Media, Manhood, and “Traditions” in a Time of Consumerism. Television & New Media. 16 (4), 331–338.

The Wire (2021) PM Modi’s ‘Mann Ki Baat’ Saw 90% Fall In Revenues in the Last Fiscal. The Wire. 20 July. [online]. Available from: https://thewire.in/government/pm-modi-mann-ki-baat-fall-revenues (Accessed 2 April 2022).

Venugopal, V. (2017) Maximum listeners tune into Mann Ki Baat from Bihar. ET: The Economic Times. 17 March. [online]. Available from: https://economictimes.indiatimes.com/news/politics-and-nation/maximum-listeners-tune-into-mann-ki-baat-from-bihar/articleshow/57678794.cms (Accessed 2 April 2022).

Waisbord, S. & Amado, A. (2017) Populist communication by digital means: presidential Twitter in Latin America. Information, Communication & Society. 20 (9), 1330–1346.

Wierzchoń, S. & Kłopotek, M. (2018) Modern Algorithms of Cluster Analysis. Studies in Big Data. Vol. 34. Cham: Springer International Publishing.

Wojcik, S. & Hughes, A. (2019) Sizing Up Twitter Users. p.1:23. [online]. Available from: https://www.pewresearch.org/internet/2019/04/24/sizing-up-twitter-users/ (Accessed 2 April 2022).

Wojczewski, T. (2020) Populism, Hindu Nationalism, and Foreign Policy in India: The Politics of Representing “the People”. International Studies Review. 22 (3), 396–422.

Yan, X. et al. (2013) ‘A biterm topic model for short texts’, in Proceedings of the 22nd international conference on World Wide Web – WWW ’13. [Online]. 2013 Rio de Janeiro, Brazil: ACM Press. pp. 1445–1456. [online]. Available from: http://dl.acm.org/citation.cfm?doid=2488388.2488514 (Accessed 2 April 2022).

Appendices

Appendix 0

Figure	Title	Method	Description
1(a)	Corresponding analysis of the 200 top lemmas of Modi according to address formats	Correspondence Analysis	Correspondence analysis shows the relative relationship between two entities. This figure shows how likely are different words (displayed in blue) to appear in different addresses by Modi (in red). The closer the word (in blue) to an address (in red), higher the likelihood of it appearing in that address compared to another word that is further away. For example, the word ‘visit’ is likely to appear more in tweets, as compared to other addresses.
1(b)	Right-side focus, 1000 lemmas specification	Ibid.	Ibid.
2	Classes in the Modi corpus and contributions of various address formats	Cluster analysis	Cluster analysis groups a set of objects such that objects in the same group are more likely to appear together in natural text. Seven clusters were formed for this analysis. Words that contribute towards each of these clusters are shown in the table on the right. The heatmap on the left shows the likelihood of each cluster appearing in the various address formats of Narendra Modi. The lighter the color, the higher the chances of a clusters’ occurrence in the address format.
3	Most frequent lemmas per topic (x) and how specific to each topic they are (y) in Tweets by Modi (left) and Trump (right)	Relative frequencies and topic modelling	The figure shows a comparison between the relative frequencies of words (or lemmas) in Modi’s tweets and Trump’s tweets. Words represented in the figure are segregated by ‘topics’ (represented by different colors). A topic (like clustering in Figure 3) is a group of words that best characterizes a topic’s theme when read together. For example, the words ‘football’, ‘cricket’ and ‘rugby’ would characterize a topic about sports. The X-axis here shows the frequency of a word within a topic. For example, ‘India’ was used about 500 times in topic (Red), 1000 times inn topic (Blue) and 1,250 times in topic (Green). The Y-axis here shows the ratio of the frequency of a word in a topic and the total frequency of the word in our corpus. For instance, a word at the extreme top right indicates that is used many times compared to other words but only in one context (or topic). For example, consider the case of ‘India’ for Modi. The word appears four times because it is used in all four topics (four different contexts). It’s position on the Y-axis however tells us the relative importance it holds for each topic. India is used way more in the context of (green topic) as compared to (purple topic). Instead, what is more important for the (purple topic) is the word ‘peopl(e)’.
4	Specificity scores of top overrepresented (left) and underrepresented (right) words in Modi’s tweets as compared to other types of addresses in the Modi corpus	Specificity scores	The figure shows how over-represented or under-represented a certain word is in Modi’s different addresses. The methodology also allows for a direct comparison. For example, the words to the left of the black separator line are relative over-represented in Modi’s tweets, especially compared to his speeches. On the other hands, words to the right of this line occur more in Modi’s speeches but do not occur so much in his tweets.
5	Similarity of Modi addresses with other political and administrative locutors leaning towards populism (left) and technocracy (right)	Cosine distances	The figure shows how similar or dissimilar are Modi’s different addresses to the language and rhetoric of other Indian locutors and documents. The extremes go from the locution being either bureaucratic in style (with many technical words and jargon) or speaking ‘saint-like’ akin to a populist (with simple, easily and widely understandable vocabulary). The black line corresponds to Modi’s tweets, while the colored dotted lines correspond to his other addresses. A value closer to 1 suggests a high similarity.
6	Specificity scores of mentions of Indian national figures in Modi’s various address formats	Specificity scores	The figure shows how over-represented or under-represented are Indian national figures in Modi’s different addresses. For example, the mention of ‘Gandhi’ (shown with a yellow bar) is over-represented in Modi’s tweets as compared to his speeches. On the other hand, mention of ‘Ambedkar’ is almost the same in his tweets and Mann ki baats.
7	Contrasting vocabulary in Modi’s tweets and Mann ki baats	Specificity scores	The figure shows words that are relatively over-represented in Modi’s Mann ki baats but under-represented in his tweets.