Criminal Politicians and Crimes against Women in India
Lara Jacoby* January 8, 2021 University of Groningen
Master’s Thesis Economics (EBM877A20) Faculty of Economics and Business
Supervisor: Dr. A. Postepska Co-assessor: Dr. S. Homroy
Abstract
The criminality of politicians is an important concern in India, the world’s largest democracy, especially since the characteristics and the quality of political leaders are assumed to influence the behaviour of their nationals. This thesis employs data on the criminal background of Indian Members of the Legislative Assembly (MLAs) to analyse the link between criminal electoral winners and the number of crimes committed against women in the corresponding constituencies. The empirical results show that there is no statistically significant link between politicians with pending criminal charges and the number of gender-based violence incidents. Evidence is provided that the variations in crime occurrences between the states included in this analysis are rather driven by unobserved state heterogeneity in the data.
JEL classifications: D72, J12, J16, K42
Keywords: politicians, crime, gender crimes, elections, India * E-mail address: l.jacoby@student.rug.nl
1
Introduction
Gaining independence from the British Commonwealth in 1947 led the Republic of India to become the largest democracy in the world, comprising over 900 million eligible voters eagerly casting their vote throughout elections.1 While the transition from a country that has long been defined by colonialism, partition, hesitant economic growth, poverty, and low education and literacy levels to a federal parliamentary democracy was fairly successful, India has been facing a rather considerable issue: there has been a steady rise in politicians with a criminal background across different kinds of elections. In 2019, 44 percent of elected Members of Parliament (MPs) in India were officially accused of committing some sort of crime, whereas one in three MPs faced severe criminal accusations.2
The election of politicians with a criminal background is a known issue both in India and at the international level and raises the question as to what extent political parties can nominate criminally accused politicians as candidates so that democracy will not be threatened. It is widely known that selecting trustful political leaders and thereby preventing the abuse of power in high positions is essential for the adequate administration and governance of a country. The latter holds all the more for developing countries where representatives are facing fewer constraints imposed by supervising institutions (Caselli and Morelli, 2004; Besley and Reynal-Querol, 2011; Dal Bó et al., 2017), thereby often inducing a strong negative impact on welfare (Chemin, 2012).
What is all the more concerning is that many of these accusations comprise crimes committed against women; an incident that has been witnessed on a vast scale in India. This violence against female members of society is regarded as being indicative of gendered differences in authority and control encouraged by strongly rooted views and customs of patriarchy. Despite the implementation of several laws to prevent these kinds of crimes, many of them can be gotten away without any form of punishment. The reasons for this are likely to be the rather oppressive position of men in the Indian society and women being regarded as the sexual property of their husband, circumstances that stay widely uncontested (Gangoli and Rew, 2018).
India is, in that sense, an example of why democracy means little when the rule of law is lacking and of why it cannot be used as a synonym for good government or
freedom, as is often done so; it merely means that the majority rules. And if the majority of people and even public leaders pursue values that do not condemn and even enforce misogyny and gendered violence, crimes against women are likely to increase and become even crueler. The country has been making progress by initiating amendments regarding gender-related violence and the oppression of minorities to their criminal law, by implementing policies to enforce female representation and their voices, as well as by altering perceptions on the victimisation of women that experienced sexual offences. Yet, the large and even increasing number of elected representatives with a criminal background, which often relates to gendered crimes, leaves little room for hope of a quick transition to a more secure environment for women and a change in perceptions among Indians. The alarming situation à propos Indian nationals’ still highly prevalent ignorance of women’s issues, of their deprivation of an elementary human right, and of how their oppression, abuse, and discrimination have vastly negative effects on a number of spheres – such as women’s mental and physical health and their quality of life but also the pace of poverty reduction, economic growth, and human capital transmission to the next generation – makes it even more vital to once more create an opportunity to shed light on the matter and its weight. Especially, since the number of gendered crimes has steadily been on the rise in India.
Electing criminal MLAs may therefore be a sign for a higher prevalence of issues at the constituency level, including violence, criminality, and lower well-being. Thus, studying local elections may be more perceptive of an association between elected criminal politicians and violence against women.
Besides that, even though several studies exist that deal with the criminality and corruption of, inter alia, Indian politicians (Pande, 2008; Aidt et al., 2011; Tiwari, 2014; Banerjee et al., 2014; Nanda and Pareek, 2016; Gehring et al, 2019; Prakash et al., 2019), to our knowledge, there are few to inexistent actual estimates of the magnitude and the social consequences of electing politicians with criminal accusations, especially when it comes to violence against women. Particularly, there are limitations on the empirical evidence of a relationship between the election of criminal politicians and the number of gendered crimes committed in places where these politicians are elected. This leaves open the question of whether there is a direct negative causal effect from electing criminal politicians onto gendered violence incidences, or whether committing these kinds of crimes is merely a deeply rooted practice in the Indian society originating from their peculiar social norms. We expect the fact of having a criminal win the legislative assembly elections in a constituency to be detrimental to the number of offences perpetrated against women in that constituency since research shows that political leaders promote crimes against women as well as felonies that are comparable to their own (Chemin, 2012), which are oftentimes offences against the human body, as can be seen by looking at the publicly and readily available election results all across India.3
Data on the number of crimes committed by Indian politicians and unique records of the prevalence of violence towards women in districts comprising the states included in this analysis are used to study the variations in gender-based offences at the constituency level once a politician with(out) criminal records is elected. In response to the ever-rising criminality among Indian politicians and the hints that having unlawful representatives at power positions provokes citizens to increasingly resort to violence as well, this study examines this association employing data on four elections in distinct major states in India, namely Bihar, Maharashtra, Tamil Nadu, and Uttar Pradesh.
The findings of this thesis show that there is no direct association of politicians’ criminality and crimes against women in our data, but that the variations in crimes against women are rather driven by unobserved state characteristics. This finding is
robust through different specifications, indicating that controlling for state-specific effects is primordial to get valid outcomes when studying the relationship between politicians’ criminal background and gender-based violence in India.
The structure of the analysis is as follows. Section 2 provides an overview of the existing literature addressing the topic of criminal politicians, crimes against women, and the relevance and influenceability of social norms in relation to gender roles and electoral processes in India. In section 3, the methodology is presented whilst section 4 follows up with the results of the analysis. In the fifth section, the findings and limitations of this study are discussed. Finally, conclusions are drawn and policy implications are given in section 6.
2
Literature review
This thesis ties in with the literature about the criminalisation of the electoral and political system, the prevalence of violence against women, and the particular relevance and influenceability of social norms in relation to gender roles and electoral processes in India.
2.1 Criminal politicians in Indian elections
The first strand of literature focuses on the criminalisation of politics, and particularly on the reasons why criminal politicians are put up for election and increasingly voted for in India. The personal background of politicians, especially concerning their corruption and criminality, are issues that are widespread among developing and transitioning countries (Mauro, 1995; Svensson, 2005; Khemani et al., 2016), generally involving weak institutions, exorbitant bureaucracy, as well as social standards tolerating the existence of apparent corruption (Nanda and Pareek, 2016). The reasons for developing countries being more prone to having unlawful representatives at powerful positions may be the higher prevalence of low income levels and low human capital stock, less economic openness, and more regulations as to market entry and press freedom in these countries (Svensson, 2005).
matter. This is particularly striking because the country enjoys notable transparency and a high participation rate when it comes to elections which would rather suggest low vote shares, as well as low contesting rates for criminal candidates. More precisely, political candidates are forced by a landmark judgement amended by the Supreme Court of India in 2002 to submit a legal affidavit divulging their criminal accusations and convictions, their education, and their financial assets and liabilities. In addition to this, the Court mentioned that the media should give particular attention to the content of these affidavits for voters to be able to take into account this novel information when voting for MPs or MLAs. After the implementation of this judgement, one would have thus expected fewer criminal politicians to be elected since voters are publicly made aware of the personal history of politicians and the criminal charges held against them. Yet, there has been a steady increase in elected representatives with a criminal background (Vaishnav, 2017). Looking at Lok Sabha elections, while in 2009, 30 percent4 of winners had declared criminal cases, the share of winners with criminal records rose to 34 percent in 2014 and to 43 percent in 2019 (Association for Democratic Reforms, 2014a, 2014b, 2019). And although some may think that the distribution of politicians with criminal accusations is intrinsic to some regions only, a high share of criminally charged politicians being elected into the lower house of Parliament partly shows that the geographic scope of criminal politicians extends all across India. According to Vaishnav (2017), this is a sign that the local prevalence of corruption, poverty, and retardation in development is not a valid argument that would explain why Indian nationals vote for politicians with pending criminal cases to such an extent.
More specifically, as noted by Banerjee et al. (2014) and Vaishnav (2017), Indian citizens tend not to be averse to voting for and even re-electing criminal politicians despite knowing about their background. The reasons for this are likely to be religion and caste conflicts, ethnic and social division, statal failure, and the lack of efficient government institutions, leading to an incapacity of supplying elemental public goods and services. When honest candidates are not able to be credible candidates and to fulfil basic requirements claimed by their people, the latter being dissatisfied with their situation turn to alternatives (Banerjee et al., 2014; Vaishnav, 2017), those being candidates with criminal antecedents in the case of India. That is to say, the use of violence and having broken the law are signs of them being able to circumvent the
The explanation for why political candidates with criminal antecedents are highly prevalent in powerful positions and managed to enter politics in India in the first place can mainly be traced back to monetary reasons. Elections and the quest for votes through election campaigns require the existence of substantial financial resources that seem to be ever-increasing. Taking the statistics from 2019, a total of 8,040 candidates contested the Indian general elections.5 Given this high number of electoral candidates, it is merely impossible for political parties to fund all the individual election campaigns and to rely on corporate donations only, leaving the electoral contestants with the duty to single-handedly back their run to become an MP or an MLA. Accordingly, less affluent, honest candidates determined to win started resorting to illicit means in order to fund their campaigns themselves. They mainly reverted to donations made by criminals, since the latter tend to possess greater funds or to gain access to liquid cash more easily through outlawed commercial activities, patronage networks, or coercion. When these sponsoring criminals understood that elections were mainly being won due to their funds, meaning that they became aware of their “money power”, they eventually made the decision to enter politics themselves (Vaishnav, 2017). As a matter of fact, Asher and Novosad (2020) show that, as rents from natural resources increase, criminal officials manage to yield private enrichment as well. Another asset of criminal candidates is that they have a reputation for being more inclined to invest their resources in political purposes. Thus, they are able to autonomously finance their elections while all the same increasing their party’s cash reserve, thereby subsidising contesting candidates of the same party that are less wealthy. This made them become the primarily chosen candidates by parties to put up for election (Berenschot, 2011; Tiwari, 2014; Vaishnav, 2011b, 2017), a phenomenon that is not expected to vanish in the near future since voters seem neither to be able nor to want to penalise political parties that appoint candidates with a tainted personal background. In that way, the appeal of putting up criminal candidates that are more likely to win offsets the incentive of nominating only candidates of higher quality which would, in other cases, leave a better impression of the political party (Tiwari, 2014). Dal Bó and Finan (2018) propose an alternative cost-related approach for why politicians with lower qualities, or in this case with criminal antecedents, may dominantly run for elections. These authors postulate that, if the costs of running as an electoral candidate rapidly increase with the quality of a politician, i.e.
their competence and integrity, it is rather the low-quality types of politicians that are nominated in elections. The reason for this is that higher-quality politicians hold increased competences that would give way to higher revenues in the labour market which are eventually forgone by being put up for election. Given that the likelihood of being elected is low and oftentimes unpredictable (Kashyap, 2007; Vaishnav, 2017) and costs to run for election are already high due to excessive electoral campaign expenses in India, this would be a complementary explanation for why individuals with a tainted past are particularly prevalent candidates in Indian elections.
Adding to the previous argument, another reason for their entering politics directly is to ensure political protection from state scrutiny instead of having to negotiate arrangements with different contesting candidates or parties, since strong political competition implies high uncertainty as to who wins an election (Kashyap, 2007; Vaishnav, 2017). By contesting the elections themselves, criminal politicians need not enter negotiations but can take advantage of their position to undermine the prosecution and sanctioning of black market activities and other illicit undertakings they are involved in (Berenschot, 2011; Acemoglu et al., 2013; Vaishnav, 2017). Thus, running as electoral candidates represents a substitute for lobbying for control, protection, and power. According to calculations done by Vaishnav (2017), unlawful candidates demonstrate a likelihood to win elections that is more than twice as high compared to that of candidates without pending criminal cases. This may partly be due to them making use of their connivance and their “muscle power” during elections, that is, using violence and capturing booths (Verma, 2005), or resorting to blackmailing and elector coercion (Aidt et al., 2011; Gehring et al., 2019). Under strong electoral competition, criminal politicians have turned into the preferred candidates set up by political parties (Aidt et al., 2011; Tiwari, 2014). They have eventually become the norm rather than the exception in India.
2.2 Crimes against women in India
The second strand of literature considers the position of women in India and the resulting crimes committed against females. The term “violence against women” includes numerous types of violence, such as intimate partner violence, assault, rape, dowry deaths, and the trafficking of women. Incidents of offences against women, especially sexual ones, are not intrinsic to some particular economic or political framework, nor can they only be observed in certain strata of society or regions in the world. The prevalence of gendered violence is spread across all wealth levels, as well as different races and cultures, and is mostly an outcome of long-established ethnic norms and beliefs. In many countries, some deeply rooted values are still being sustained through institutions in political, economic, and societal domains and therefore foster the oppression, discrimination, and the subordination of women (World Health Organization, 2013). Yet, crimes against women are renowned to be a prevailing and serious problem in India in particular due to persistent patriarchal control and to wide-ranging societal inequality which is strongly related to violent crimes (Morgan, 2000), often targeted at women. According to the National Crime Records Bureau (2020), there has been an increase in occurrences of crimes against women of 5.1 and 7.6 percent in the years 2018 and 2019, respectively. While especially intimate partner violence and offences committed by relatives or acquaintances have long been swept under the carpet, a promptly growing research corpus has emerged due to better measurements of female exposure to all sorts of violence by means of household and specialised national surveys. In addition to improved quantification of gendered violence, more focus has generally been put on prioritising women’s safety and on improving the availability of services for women that want to seek help (World Health Organization, 2001), thereby possibly increasing female reporting of abuse and violence towards them in India. This movement of female bravery is partly due to women in developing countries not accepting the customs of child marriage and of their male counterparts resorting to domestic violence anymore, indicating the aversion of women to gender-specific roles and conduct in countries where gender disparities are substantially prevailing (Asadullah et al., 2020).
2002). In accordance with these norms, gendered violence at the household level seems to be the standard as well as broadly tolerated, given that the reasons for resorting to violence, such as wives’ disobedience or their infidelity, are regarded as being entirely valid motives to do so (Rao, 1997; National Commission for Women, 2000). What is even more concerning is that the existence of domestic violence is generally denied due to conservative family culture where mentioning sexual intercourse is seen as being out of bounds, leading to serious underreporting of incidents of intimate partner violence to the police (International Institute for Population Sciences (IIPS) and Macro International, 2007; Gupta, 2014). Palermo et al. (2013) find that, for the years 2005 and 2006, only 32 percent of women proceeded to any kind of report of gender-based violence, while merely one percent of these reports were formal reports to either the police, or to medical or social services. Police forces have traditionally been unhelpful when it comes to reports of domestic violence since the sanctity of the family was and still is an important and traditional value in India (Bush, 1992), and since the speed of investigation into gendered violence occurrences is low, inducing weak confidence in the law enforcement bodies (Mandal, 2019). The justice system is shown not to provide a hospitable environment either for women to feel like they are taken seriously by reporting any kind of offence against them; it rather represents a barrier. This, since the processing of reports of gendered violence usually comes with negative connotations for victims, such as humiliation, degradation, and victim-blaming (Banarjee, 2020). Further, there has been a decline in condemnation rates, especially for perpetrators accused of dowry deaths and rape. The existence of insufficient convictions for rape is presumably the case because it is difficult for women to provide solid evidence of actually having experienced rape and to gain overall credibility. This induces a loss of faith in the judicial system concerning the initiation of effective judicial proceedings against charged offenders and therefore to retention when it comes to reports of violence against women, especially for crimes with low conviction rates (Mandal, 2019). Regarding dowry deaths, the reasons for a low likelihood of offenders being convicted are questionable since lethal crimes are difficult to conceal and evidence can more easily be provided such that the perpetrator could readily be held accountable (Drèze and Khera, 2000).
crimes. A violence-free environment for women would mean, at one hand, a higher quality of life for them (Sen, 1985) and the acquisition of an elementary human right, while all the same bringing about socioeconomic advantages, such as lower fertility and reduced child mortality (Eswaran, 2002). Further, it could trigger economic growth and improve the performance of public institutions (Sen, 1985), as well as prompt female education and employment achievements and participation (Morton et al., 2014), thereby reducing poverty and facilitating the transmission of human capital to the next generation (Rosenzweig and Schultz, 1982). Violence against women is therefore not only an individual issue but also and foremost a trigger for more serious governmental intervention (Drèze and Khera, 2000; World Health Organization, 2001).
note, Priyanka (2020) studies the long-term economic effects of women being elected to legislative assemblies in India and finds that young female adults derive substantial benefits in terms of an increased likelihood of participating in the labour market and of completing higher levels of education, possibly due to altered aspirations through a role model channel. These results show that putting more women in powerful positions is a viable means to achieving higher reporting of gendered violence, whilst all the same being in line with other studies that highlight the importance the identity of politicians has on policy outcomes, interventions, and societal evolution (Besley and Coate, 1997; Pande, 2003; Chattopadhyay and Duflo, 2004; Bhalotra et al., 2012; Meyersson, 2014). As can be seen when looking at the criminal records of election winners,6 a larger number of politicians’ criminal cases are minor charges or associated with candidates’ political activity. Having said that, many criminal politicians are accused of offences against the human body. In the Indian Penal Code (IPC), the said category includes all sorts of crimes committed with regard to violence towards women, such as kidnapping, dowry deaths, and rape; incidents that are, as already claimed, not uncommon in India and that have gained increased international attention due to the occurrence of the infamous 2012 Nirbhaya gang rape in Delhi. The inadequate response of the Indian government regarding this outrage and the related public revolt unveiled and underlined patriarchal ideologies and misogyny, adverse ideas of shame and dishonour in the Indian society, questions about women’s safety, and hegemonic handling of sexual violence such as victim-blaming (Lodhia, 2015). As a response to the public outcry following the
Nirbhaya gang rape and to the atrocity of the act, the Indian Government assembled a
committee to go over criminal law acts and suggest modifications, to put forward procedures that allow for quicker trials, and to advance stricter sanctions for sexual crimes. The resulting document, the Verma Committee report, thereby represents a decisive point concerning the handling of crimes against women. It predominantly stresses the responsibility of governmental institutions in condemning gender-based violence and in criticising the way in which crimes against women are treated at the legal and conceptual level. Other aspects of the report mention the relationship between violence towards women and gender inequity, claim the right for protection, and declare that sexual offences are the result of noxious patriarchism (Verma et al., 2013). The acknowledgements and propositions stipulated in this report were nevertheless only
partly incorporated in the Criminal Law (Amendment) Act (2013). In that regard, the, for India, unprecedented recommendations of the Verma Committee mostly regarding marital rape and violence were not sufficiently enforced in the new law. A couple of other legislations promoting women’s protection, such as the Dowry Prohibition Act (1961), the Indecent Representation of Women Prohibition Act (1986), the Protection of Women from Domestic Violence Act (2005), and the Sexual Harassment of Women at Workplace (Prevention, Prohibition and Redressal) Act (2013), have been implemented over the course of the years in India. Yet, no significant improvements in terms of reduced crime occurrences have been found – rather the opposite. Even though the increase in incidents is partly due to more reporting of especially domestic violence by female victims, male culpability, as well as the awareness of violent masculinity and of gender inequality are still considerably lacking among Indian nationals.
2.3 Importance and influenceability of social norms in the Indian society
men to resort to increased violence against women to try to preserve the conventional male-female hierarchy (Bradley, 2009; Staudt, 2011; Neuman, 2013). In accordance with this idea, Kishor and Johnson (2004) detect a positive association between women’s employment and partner violence. Chin (2012) complementarily suggests that violence against women, linked to increased labour force participation, might be the result of women possessing higher monetary resources, especially in ethnic contexts where patriarchy is strongly prevalent, husbands then tending to proceed to backlashing and extracting their wives’ income to restore dominance. The fear of recurring violence against them, both in a domestic setting and due to increased crime rates, may then result in a low female participation rate in the labour market (Chakraborty et al., 2018). While India is not the only nation where gendered crimes have significantly increased, the counterproductive implications of its economic progress preoccupy many researchers. Economic development usually implies a slow but steady change in social and cultural values, generally towards more modern mentalities since new ideas of social equity and new gender roles are then introduced to society (Trivedi, 2010; Neuman, 2013). In India, economic growth occurred rather quickly so that social transformations could not keep up with its pace, thereby leaving the new role of women in society opaque and hardly comprehensible for their male counterparts. Violent behaviour may then simply be an inept reaction to women embracing more advanced attitudes towards their integrity, independence, and their rights (Eswaran and Malhotra, 2009; Simister and Mehta, 2010; Hackett, 2011), and to women breaching the prevailing social norms that begin not to be taken as given anymore by them. In contrast to the situation in India, various international studies postulate that the positive relationship between economic expansion and crime rates is arguably rather due to increased reports of violent incidents than to an actual increase of offences (Mandal, 2019).
the effect of electing a criminal political candidate on illicit acts and finds that there is an increase in the general level of crimes and in crimes against women. Criminal political leaders are further found to promote crimes that are comparable to their own which is alarming for the situation of women in India. The increased election of criminal politicians and thereby the wide-ranging spread of their personal beliefs regarding the condition of women and other minorities in India may then provoke yet more detrimental effects for these groups. This, since the quality of the political sphere and of the societal well-being is to a certain degree determined by individuals who exhibit questionable values and norms that are reflected in their own criminal behaviour. All the more, Besley et al. (2005) show that not only can politicians affect voter behaviour, the targeting of recipients of certain programs, and the values that are still supported among certain groups of society, but also do the socioeconomic characteristics of the villages the candidates come from have an influence on political selection through their quality and on incumbents’ conduct whilst in power, often related to the allocation of resources for beneficiary purposes. This kind of reverse causality implies that concentrating on elements that allow for higher-quality politicians to be elected in the presence of group identity constraints is central to enhancing governmental performance and thereby its institutions.
well (Aragonès et al., 2020). Accordingly, Banerjee and Pande (2007) show that constituencies comprising a greater number of low-caste electors more frequently display political winners with a criminal background. These populations are found to mostly be poor and easily intimidated, as well as to having low literacy levels so that organised gangs are more likely to fruitfully use their muscle power for dishonest electoral candidates to be re-elected (Wilkinson, 2004). Regarding the fact of certain parts of the population being illiterate, regions with a higher prevalence of illiterate eligible voters, overlapping with areas where more electors come from lower castes, demonstrate a larger share of candidates with a dubious past (Aidt et al., 2011).
3
Research methodology
3.1 Data
The data consist of individuals who are elected as MLAs, otherwise known as members of the Vidhan Sabha. Eligible voters in a state directly cast their votes for contestants through secret ballots, where each constituency can only have one electoral winner who is determined by a simple majority vote victory. The elected legislators are responsible for making laws for their respective states, they can collect and impose taxes, and are involved in electing the President of India as well as the representatives of their state in the upper house of Parliament. Existing literature shows that local governments benefit from extensive authority over the selection of governmental welfare programs that can be utilised in their favour (Matthew and Buch, 2000), and enjoy executive responsibilities, notable informal power, and control over administrative assignments so that they can decide over the formulation and the enforcement of new policies (Prakash et al., 2019).
We analyse legislative elections in four different states, the states of interest being Bihar, Maharashtra, Tamil Nadu, and Uttar Pradesh with the period ranging from 2009 until 2012. Concerning location, Tamil Nadu is situated in the south of India and Maharashtra lies further to the west, while Bihar and Uttar Pradesh are neighbouring states in the north of the country bordering on Nepal (see Figure A.1. in the appendix). For each of these states, one specific assembly election is focused on: we consider the elections of 2009 for Maharashtra, for Bihar those of 2010, and concerning Tamil Nadu and Uttar Pradesh we study the 2011 and 2012 elections, respectively. According to the Indian constitution, these elections are scheduled elections that occur every five years and for which the timing is anticipated. Thus, they are not rolled out to hastily implement new government policies wanting to tackle criminality issues across states for instance and can be perfectly anticipated by potential candidates. In that sense, incumbents and contesting candidates can deliberately prepare for the elections by trying either to improve their image or to put their rivals in a poor light.
Details of electoral results are gathered from Myneta,7 a website providing information about elections and the contesting candidates to Indian voters, and run by the Association for Democratic Reforms (ADR). The ADR is a non-governmental
organisation that is concerned with reforms in the political and the electoral sphere. Its main goal is to achieve more transparency in politics in India, as well as to limit the impact of money and muscle power in elections. Myneta thus serves to supply reliable and thorough background information on electoral candidates and on several political parties, mainly retrieved from affidavits that individuals have to provide themselves to be accepted as an electoral candidate. We make use of this website to fetch specific data about legislative election winners regarding the four elections of interest in this study, i.e. the party they belong to, their age, education, total assets, and the number of pending criminal cases they hold. Concerning the assets of candidates, they comprehend movable and immovable assets of the candidates themselves, as well as of their spouse and their dependents. This information from Myneta allows us to include a total of more than 1,100 electoral politicians in the analysis of criminal politicians in India.
In this study, two measures of the criminality of electoral winners are created. The first one is a binary variable crimicases indicating whether the winning candidate has a criminal background. The second one, crimcount is an integer comprising the number of pending criminal cases held by a candidate. More specifically regarding the criminal background of politicians, the emphasis in this study lies on criminal accusations rather than convictions. Since it can oftentimes take years, if not decades until pending cases are treated and politicians eventually convicted, not possessing any records of criminal convictions does not necessarily imply that politicians have not hitherto engaged in any criminal activities. That is why this study makes use of criminal complaints instead and considers winners as being criminals for which there is prima facie evidence of being involved in any kind of criminal activity.
Unique and detailed data on the different types of gender-based violence incidents and the amount committed of each of them in several districts of the four states are retrieved from the National Sample Survey of India (NSS). Information on macroeconomic indicators at the state level, comprising the GDP and the unemployment rate, is taken from statistics provided by the Government of India. The GDP is expressed in Indian rupees, while the state unemployment rate represents the number of unemployed individuals per 1000 inhabitants. The number of eligible voters per constituency is retrieved from the Election Commission of India.
concerning the kinds of crimes committed against women can only be found at the district level. Evidence shows that constituencies are under all circumstances smaller than districts, hence the constituencies which appear in the crime data are matched to the corresponding districts using different electronic governmental sources which give insights on the subdivisions of the existing districts into constituencies, villages, and towns. This matching is, in some cases, only an approximation since political constituencies do not map one to one to districts and since the composition of districts changes over the years. Given the fact that the data do not include information on criminal acts for all, as of 2011, 640 districts in India8 and that the division into districts and constituencies has been consistently altered over the years, some constituencies and therefore some politicians are not assigned any data regarding criminal acts committed against women in several districts. The latter in no way implies that there have not actually been cases of offences against women, it merely entails missing information and thereby missing values for the analysis. Additionally removing the electoral winners for which there is no clear information on their educational level, the final sample used in this study comprises 802 politicians from four major Indian states to which can be matched a specific number of types of crimes committed against women at the district level in the respective state.
More precisely, the different types of crimes committed against women in districts of either Bihar, Maharashtra, Tamil Nadu, or Uttar Pradesh we consider in this analysis are the following: rape of women and girls, kidnapping and abduction of women, dowry deaths, assault on women with intent to outrage their modesty, insult to the modesty of women, and cruelty exerted by a woman’s husband or his relatives. To obtain constituency-wise variations in crime occurrences, merely observable at the district level, all while accounting for differences in population densities between constituencies, crimes are expressed as the number of gendered violence incidents per 100,000 eligible constituent voters. Representing these types of crimes per 100,000 eligible voters and hence by a measure that takes into account the number of main potential delinquents implies the assumption that the number of crimes is proportional to the voter population of the constituencies and that culture is the same within districts. As seen in section 2, India is a multicultural country often divided into the north and the south in analyses since these regions are found to exhibit different views in terms of gender roles and
levels of patriarchal control. Controlling for certain state variables allows us to determine whether living in different states has an effect on violence targeted at women while assuming that the fact of living in a certain state is indicative of exhibiting state-specific cultural values, and of differing tendencies to resort to violence and to elect criminal politicians.
Following Tiwari (2014), affiliation to either the Congress Party (INC) or the Bharatiya Janata Party (BJP) is controlled for which are the two major and most known parties in India, as well as the ones with the second and third-most declared criminal cases in phase one of the 2019 Lok Sabha elections,9 after the Yuvajana Sramika Rythu Congress Party (YSR Congress Party). While the INC is estimated to include around 20 million members,10 the president of the BJP announced in 2019 that the party has managed to assemble a total of 180 million affiliates.11 Because of their dimension and their popularity, but also because of the large share of criminal politicians they comprise, it is expected that the affiliation of winning MLAs to either party may be positively related to the number of crimes committed in the corresponding constituencies.
3.2 Summary statistics
The summary statistics of electoral winners’ characteristics, of the election outcomes regarding the criminality of the winners, and of the types of crimes committed against women in the different constituencies where the elections took place are presented in Tables 1 and 2. The raw data show that the majority of elected politicians in the four states in question are aged between 46 and 55 years, while the mean age is 49.5 years. The total amount of assets for the greater part of winners ranges from 25,000 to 5,000,000 Indian rupees whilst a quarter of the winners own more than 25 million rupees, the mean amount of assets rather oscillating at an even higher level of 30.1 million rupees which suggests that a fair number of politicians possess great amounts of wealth. The majority of elected representatives hold a Bachelor’s degree and 26 percent of all candidates are affiliated to either the BJP or the INC. The average number of criminal accusations for the whole final sample equals almost 2 cases, where the maximum
9 See https://www.statista.com/statistics/993869/election-candidates-share-with-criminal-records-by-party/ 10 See https://www.britannica.com/topic/Indian-National-Congress/Policy-and-structure
https://timesofindia.indiatimes.com/india/bjp-to-add-7-crore-new-members-j-p-number of accusations surges to 36. Yet, the share of politicians with no recorded criminal cases equates to about 51 percent. For the four elections in the different states, slightly more than half of the winners do not hold any criminal accusations.
Looking at the individual states, results show that the share of winners without pending criminal cases diverges between 40.21 percent for Bihar and 69.14 percent for Tamil Nadu. Bihar having such a low rate of non-criminal winners is partly not surprising since a study by Transparency International India (2005) reports Bihar as the most corrupt state in India, corruption and criminality mostly going hand-in-hand. The distribution of MLAs with criminal accusations in our data is similar to the one identified by Vaishnav (2017) who analyses elections before and during 2015. As can be seen in Table 1, the four states seem to be different with respect to choosing criminal politicians and to the average number of crimes committed against women. What stands out when looking at the mean number of politicians’ pending criminal cases is that Maharashtra displays a much higher value than the remaining states. This may seem surprising to some since it challenges the popular belief that dishonest politicians predominantly pile up in Bihar and Uttar Pradesh. The percentage of elected politicians that have broken the law seems not the be perfectly related to the mean number of pending criminal cases, given that politicians from Maharashtra appear to have a higher number of criminal accusations than those from Bihar, although the latter turns out to comprise a higher share of candidates with a criminal history. Mean age and education levels are fairly similar for all four states, but politicians in Bihar seem to hold much smaller amounts of assets than the other three states, the mean in Bihar being 8 million rupees while total assets range between 34 and 41 million rupees in the other three states. The latter also holds for state GDP, which is significantly lower in Bihar than in the other three states, Maharashtra exhibiting more than a fivefold higher GDP than Bihar. This lower GDP value for Bihar cannot be traced back to any kind of negative shocks in 2010 though which could have potentially explained the large discrepancy, but is rather due to Maharashtra, Tamil Nadu, and Uttar Pradesh exhibiting the highest nominal GDPs among all 33 Indian states over the years.12
Bihar Maharashtra Tamil Nadu Uttar Pradesh All elections
Year 2010 2009 2011 2012
Number of observations 194 193 162 253 802
% of criminal winners 59.79 54.40 30.86 47.43 48.75
Number of pending criminal cases 1.876
(3.127) 2.648 (4.719) 1.043 (3.483) 1.791 (3.637) 1.867 (3.818) State GDP (in Indian rupees) 130,171 666,944 433,238 443,191 419,308
(187,731)
Types of crimes committed against women per 100,000 eligible voters
Rape (6.696) 10.943 (14.491) 20.447 (10.952) 12.574 (6.367) 9.820 (10.751) 13.205
Kidnapping (21.466) 34.118 (10.894) 12.932 (20.130) 30.596 (28.233) 42.843 (24.434) 31.061
Dowry deaths (10.646) 17.607 (3.291) 4.155 (2.630) 2.564 (6.639) 11.769 (8.906) 9.490 Assault with intent to
outrage women’s modesty (7.638) 6.796
41.821 (27.732) 28.596 (28.442) 16.834 (12.377) 22.795 (24.009)
Insult to women’s modesty (1.116) 0.2697 (13.769) 14.349 (15.715) 7.058 (0.219) 0.072 (11.421) 4.964 Cruelty by husband or his relatives (28.989) 31.656 102.534 (64.194) 32.899 (27.180) 43.850 (42.225) 52.810 (52.051) Table 1: Election outcomes regarding the criminality of winners, state GDP, and the types of crimes committed against women in the different constituencies comprising the states in our data. For the number of pending criminal cases, state GDP for all elections, and the different kinds of crimes, mean values are displayed with standard deviations in brackets.
MLAs among these four states of 4 percent, while it corresponds to 7 and 9 percent for Tamil Nadu and Uttar Pradesh, respectively, and surges to 15 percent for Bihar. This high share of female state officials in Bihar might not seem surprising since its government had already introduced a reservation of 50 percent for women in rural governing entities in 2006,13 making participation in state assembly elections more tangible and conceivable for women in the state of Bihar.
Turning to the differences between winners with(out) pending criminal cases, what prevails is that no greater differences in sociodemographic variables can be observed, except for the fact that criminal politicians seem to hold much higher amounts of total assets; the maximum being 603 million rupees for non-criminals while it surges to 1,270 million rupees for winners with criminal records. Regarding party affiliation, 23.36 percent of non-criminal elected officials belong to either the BJP or the INC whereas 29.16 percent of criminal winners are members of one of these parties. Occurrences of the different types of gendered crimes are mostly similar for both groups, apart from cruelty carried out by a woman’s husband or his relatives which is considerably more prevalent in constituencies where a criminal politician is elected.
Winners without criminal cases
(1) Winners with criminal cases (2)
Mean Std. Dev. Mean Std. Dev.
Age 50.418 11.050 48.657 9.895
Education 5.686 1.717 5.811 1.852
Total assets
(in million Indian rupees) 27.7 57.3 32.6 102
Number of pending criminal cases / / 3.829 4.734
Observations 411 391
Types of crimes committed against women per 100,000 eligible voters
Rape 13.075 11.348 13.343 10.097
Kidnapping 31.122 24.819 30.996 24.055
Dowry deaths 8.875 8.521 10.135 9.261
Assault with intent to
outrage women’s modesty 23.056 23.921 22.520 24.129
Insult to modesty 5.144 12.085 4.775 10.690
Cruelty by husband or his relatives 49.196 49.839 56.610 54.084 Table 2: Column (1) presents the summary statistics of winners’ characteristics without criminal records and the occurrence of different types of crimes committed against women per 100,000 eligible voters in the corresponding constituencies. The characteristics of elected politicians that are accused of having committed some kind of crime and the number of gendered crimes committed in the respective constituencies are shown in column (2).
13 See
T-test results show that the mean differences for winners with(out) criminal cases are not statistically significant at the one percent significance level for each crime category, while they are significant at the five percent level for dowry deaths and cruelty. Thus, there does a priori not seem to be any considerable difference in terms of occurrences of most types of violence towards women between constituencies where a politician with criminal records is elected and constituencies where the winner has no pending criminal cases, except for cruelty and dowry death incidents.
3.3 Model specification
The empirical approach makes use of the number of pending criminal cases of electoral winners to identify potential effects on the number of crimes committed against women in the constituency where the respective politicians are elected.
For every kind of crime 𝑖 committed against women in the analysis, the following model is estimated by the Ordinary Least Squares (OLS) method with robust standard errors:
𝑐𝑟𝑖𝑚𝑒𝑖𝑗𝑠 = 𝛼 + 𝛽𝑐𝑟𝑖𝑚𝑖𝑐𝑎𝑠𝑒𝑠𝑗𝑝𝑠+ 𝛿𝑿𝑗𝑝𝑠+ 𝜀𝑖𝑗𝑝𝑠 (1)
where 𝑐𝑟𝑖𝑚𝑒𝑖𝑗𝑠 represents the number of crimes 𝑖 committed against women in a district
per 100,000 eligible voters in constituency 𝑗 of state 𝑠, the regressor of interest 𝑐𝑟𝑖𝑚𝑖𝑐𝑎𝑠𝑒𝑠𝑗𝑝𝑠 being a dummy variable that takes on a value of one if a politician 𝑝 in
constituency 𝑗 has a criminal background and zero otherwise. Other characteristics 𝑿𝑗𝑝𝑠
of election winners are controlled for, those being the winning candidates’ age, their wealth, level of education, partisan affiliation, and state dummies of the constituency the politicians are elected in where the state of Bihar serves as the base category. Additional specifications also include the state GDP and unemployment rate as covariates. 𝜀𝑖𝑗𝑝𝑠 represents the error term.
candidate in our data. Thus, it is difficult to predict the sign and the magnitude of the estimates of the main regressor. Further, our estimates cannot be interpreted as the causal effect of the criminality of politicians on the number of crimes committed against women, we rather only indicate associations. This, since the regressor that captures the criminality of politicians might be endogenous due to new legislations or to amendments to laws regarding the situation of women or the severity of punishments being implemented in certain constituencies/states. These implementations are likely to have an effect both on gendered crime occurrences in these constituencies/states, as well as on the criminality of elected winners because these reforms might deter both citizens and politicians from getting involved in criminal acts of any kind. Additionally, Equations (1) and (2) omit the fact that richer constituencies offer higher wages to elected MLAs and thereby presumably attract more criminal politicians since they are found to mostly contest elections for rent-seeking activities and gaining money (Gehring et al., 2019). Yet, richer and therefore often more developed constituencies might spend more resources on welfare programs which would imply an increased well-being of their citizens, thereby possibly reducing violence incidents and the election of politicians with a tainted past. Controlling for constituency-level wealth may give more indications on the forces driving the election of criminals and the perpetration of gendered violence. Thus, the potential confoundedness of the dummy capturing the criminality of politicians suggests that there might be endogeneity issues.
An alternative specification rather employs the number of criminal cases as main regressor in order to account for the fact that the (non)criminality of politicians in itself may not be decisive as regards gender-based crimes in constituencies, but that incidents of this type of violence are rather positively linked to the reoccurrence of crimes committed by electoral winners:
𝑐𝑟𝑖𝑚𝑒𝑖𝑗𝑠 = 𝛼 + 𝛽𝑐𝑟𝑖𝑚𝑐𝑜𝑢𝑛𝑡𝑗𝑝𝑠+ 𝛿𝑿𝑗𝑝𝑠+ 𝜀𝑖𝑗𝑝𝑠 (2)
where 𝑐𝑟𝑖𝑚𝑐𝑜𝑢𝑛𝑡𝑗𝑝𝑠 denotes the number of pending criminal cases held against the state
legislative assembly winner 𝑝 in constituency 𝑗 of state 𝑠. For Equation (2), the covariates included in 𝑿𝑗𝑝𝑠 are politicians’ age, wealth, education level, and the state
4
Empirical Results
This section presents the OLS estimates of the effect of electing a (non)criminal MLA on the district-level occurrence of different types of crimes against women weighted by the number of eligible constituent voters. For each of the dependent variables mentioned in the previous section, there are four empirical models to begin with. The first parsimonious specification includes merely a dummy for whether a politician has a criminal background or not. Model 2 adds the personal characteristics of elected MLAs as covariates such as their age, their wealth, their education, and the state of the constituency where they are elected to Model 1, while in specification 3 party affiliation is additionally controlled for. The fourth specification is an estimation of Equation (2) with the same covariates as in Model 2. Regression results for Equation (2) with only the number of criminal cases as a regressor and by using the same covariates as in Model 3, respectively, are presented in Table B.1. The reason for this is that, first, the main interest of this study is to determine whether being a criminal politician or not has an impact on incidents of gender-based violence rather than to uncover a link between an increasing number of criminal accusations and the number of crimes committed against women. Second, party affiliation is found not to significantly improve the model fit, thereby making it redundant to present these estimation results in direct comparison with the specifications we are especially interested in.
Table 3 thus reports the relationship between the criminality of an electoral winner and the six different kinds of gender-based violence for four specifications. Since the regressions are estimated via OLS and there is supposed to be endogeneity of the regressors, the point estimates document basic correlations without addressing causality. Each entry in the columns for the regressor signalling either the criminality or the number of pending criminal cases of politicians represents the estimated coefficient on a specific left-hand-side variable, indicating the number of crimes committed against women, in four separate regressions.
which may not seem surprising since the t-tests showed that mean differences in crime incidents between constituencies with a (non)criminal winner are only significant for these types of crimes.
covariates generally seem not to be decisive in determining the impact of the criminality of political representatives, including the states comprising the constituencies where MLAs are elected as control variables appears to explain much of the variation in gendered violence circumstances.
Depending on the type of crime, living in a certain state significantly increases or decreases the number of criminal events. Taking the state of Bihar as the base category, assaults, insults, and cruelty are more prevalent in the other three states, a link that is statistically significant at the five percent level except for insults in Uttar Pradesh and cruelty in Tamil Nadu. Dowry deaths are significantly less common in all the states as opposed to Bihar at the one percent significance level. There is a statistically significant decrease in kidnapping incidents in Maharashtra and Tamil Nadu at the one and ten percent significance levels, respectively, but yet an increase in Uttar Pradesh which is statistically significant at one percent. Rape reports are more common in Maharashtra and Tamil Nadu than in Bihar, while the estimate is only statistically significant at the one percent significance level for the foremost state. Compared to Uttar Pradesh, Bihar shows significantly more rape incidents. Examining these estimates thus unveils the reason for not finding significant estimates for the link between the criminality coefficients and the number of crimes committed against women; in these regressions, the state dummies turn out to be the main drivers for the observed differences in gender-based violence incidents.
inherent to the separate states than to the fact of electoral winners having a criminal background in our data.
alone and from patronising dangerous areas through awareness campaigns, and increasing the sanctions for (kidnapping) offenders may be ways through which kidnapping incidents are reduced by more educated politicians.
A tendency which can be observed across the specifications that include covariates for electoral winners’ age, wealth, education, the state of the constituency where they are elected, as well as party affiliation, is that, for all crime categories, Maharashtra and Tamil Nadu seem to follow the same pattern regarding an increase or decrease of the number of crimes in comparison to the base category, i.e. Bihar. The estimates of the state of Uttar Pradesh mostly heading in the same direction as well, the signs of the coefficients for rape and kidnapping are yet opposing. An explanation for observing a similar pattern in the impact on crimes when controlling for state dummies may be that Maharashtra and Tamil Nadu are states with higher measured human development indices (HDI), while Bihar and Uttar Pradesh, apart from being neighbouring states, share very similar as well as the lowest HDIs in India for all recent years.14 Further, as regards location, Tamil Nadu and Maharashtra are states that lie rather in the south or south-west of India, commonly thought to be more open to female independence due to patriarchal values and kinship arrangements being less widespread and persistent (Dyson and Moore, 1983; Eswaran et al., 2013). This would again hint at a weighty association between specific state elements and the criminality patterns in the constituencies encompassed within the states.
Table 3: Regression results Rape (no covariates) (1) Rape (added covariates) (2) Rape (added covariates) (3) Rape (added covariates) (4) Kidnapping (no covariates) (5) Kidnapping (added covariates) (6) Kidnapping (added covariates) (7) Kidnapping (added covariates) (8) Criminality of election winners (dummy) 0.268 (0.758) (0.728) -0.063 (0.728) -0.064 (1.726) -0.127 (1.597) 0.450 (1.597) 0.448 Number of criminal cases (0.076) -0.085 (0.158) -0.213 INC or BJP affiliation 0.747 (0.875) (1.770) 1.786 Age 0.053 (0.033) (0.033) 0.052 (0.033) 0.051 0.186** (0.082) 0.185** (0.082) 0.180** (0.081)
Table 3:Regression results (continued 1) Dowry deaths (no covariates) (9) Dowry deaths (added covariates) (10) Dowry deaths (added covariates) (11) Dowry deaths (added covariates) (12) Assault (no covariates) (13) Assault (added covariates) (14) Assault (added covariates) (15) Assault (added covariates) (16) Criminality of election winners (dummy) 1.260** (0.629) -0.341 (0.49) (0.490) -0.341 (1.697) -0.537 (1.49) 0.811 (1.490) 0.810
Number of criminal cases -0.079
(0.056) (0.162) 0.026
INC or BJP affiliation -0.083
(0.598) (1.617) 1.236
Age 0.045*
(0.024) (0.024) 0.045* (0.024) 0.045* (0.067) 0.077 (0.067) 0.076 (0.068) 0.075
Note: Robust standard errors are in parentheses. *Significant at 10%; ** significant at 5%; *** significant at 1%. In total, there are 802 observations for all specifications. Regarding total assets, the reference category is 0.025 – 5 million Indian rupees, while for education the base category is “literate” and for state the omitted category is the state of Bihar.
Table 3: Regression results (continued 2)
Insult to modesty (no covariates) (17) Insult to modesty (added covariates) (18) Insult to modesty (added covariates) (19) Insult to modesty (added covariates) (20) Cruelty (no covariates) (21) Cruelty (added covariates) (22) Cruelty (added covariates) (23) Cruelty (added covariates) (24) Criminality of election winners (dummy) -0.370 (0.805) (0.699) -0.241 (0.699) -0.243 7.414** (3.678) (3.169) 3.503 (3.171) 3.504
Number of criminal cases 0.041
(0.125) (0.445) 0.336
INC or BJP affiliation 0.939
(0.742) (4.107) -0.488
Age -0.032
(0.032) (0.032) -0.033 (0.032) -0.030 (0.146) 0.136 (0.146) 0.136 (0.146) 0.131
as the reference category. The second column shows regression results when the state unemployment rate is used as a covariate instead of the state dummies in the model. Estimates for age, wealth, and education are not shown in the table since they are identical and statistically insignificant for both specifications. While the magnitude of the estimates capturing the state characteristics is different for both specifications, in every case the sign of both coefficients is identical for each type of gendered violence as well as significantly different from zero. Interestingly, the sign of the state unemployment rate variable is negative for all but one kind of crime, the exception being kidnapping. This would mean that a higher unemployment rate would lead to decreases in criminal activities, which is in contrast to most studies dealing with the link between unemployment and criminality. An explanation for this is probably that in our data, we only consider information about one year for each state, meaning that in this regression, we only compare two unemployment rate values. The signs of the unemployment rate coefficients coinciding with the ones from the state dummies and the ratio of the state unemployment rate estimates to the estimates of the state dummies being identical for each dependent variable, these results cannot be interpreted as the effect of the magnitude of the state unemployment rate on the differences in gendered crime incidents. Rather, it is again an indication of how the number of gender-based crimes diverges between the two states in this regression due to individual state characteristics.
Table 4: Regression results including either the state dummy or the state unemployment rate
Table 4: Regression results (continued) Assault (state dummy) (7) Assault (unemploy-ment rate) (8) Insult (state dummy) (9) Insult (unemploy-ment rate) (10) Cruelty (state dummy) (11) Cruelty (unemploy-ment rate) (12) Criminality of election winners (dummy) 2.30 (3.262) (3.262) 2.30 (1.654) -0.939 (1.654) -0.939 (5.732) 6.715 (5.732) 6.715 Age X X X X X X Total assets X X X X X X Education X X X X X X State Tamil Nadu -12.026*** (3.498) -7.665*** (1.77) -66.424*** (5.492) State unemployment rate -2.004*** (0.583) -1.278*** (0.295) -11.071*** (0.915) Constant 44.461 (15.473) (18.065) 76.531 29.573 (7.15) (8.108) 50.014 (19.769) 92.672 (26.986) 269.802 R-squared 0.089 0.089 0.082 0.082 0.351 0.351 Adj. R-squared 0.046 0.046 0.038 0.038 0.320 0.320
Note: Robust standard errors are in parentheses. *Significant at 10%; ** significant at 5%; *** significant at 1%. In total, there are 355 observations for both specifications. Regarding total assets, age, and education, the estimates and the statistical significance are identical for both models, as well as for the ones in Table B.1. in the appendix. For the state dummy, the base category is Maharashtra. The state unemployment rate is expressed as the unemployment rate per 1,000 inhabitants.
Apart from kidnapping, Tamil Nadu appears to significantly have lower numbers of incidents of crimes against women than Maharashtra. The estimates regarding the criminality dummy are identical for both specifications as well, yet being statistically insignificant while the magnitudes and signs of the estimates are diverging for the different types of crimes. These specifications provide some more evidence that the variations in the number of crimes committed against women are primarily driven by the state dummies and thus the individual state characteristics rather than by the criminality of politicians in our data. Comparing only Tamil Nadu and Maharashtra, two of the states with the highest as well as very similar HDIs and GDPs in India, makes clear that, even when taking into account these indicators, there are dissimilarities between states not directly related to (socio)economic conditions which induce the states to be different. The results including the state GDPs are presented in Table B.2. in the appendix since the differences in GDP levels between both states are such that the state GDP estimates are close to zero but statistically significant at the one percent level.
development has an impact on the occurrence of domestic violence in India, such that more developed states show reduced incidents of dowry deaths. Yet, regarding wife maltreatment, development alone does not suffice to prompt a reduction of incidents since the course and pace of development are of importance as well. Especially in cities of states that are less developed and in the rural regions of states that have experienced more social development, questioning long-established gender roles may even result in more domestic violence. This again suggests that there are other drivers, likely to be customary values and beliefs regarding the role of the woman in society, as well as traditional predilections to resort to violence towards women prevailing in the separate states, that account for the variations in gendered violence incidents across Indian states. It might yet be possible that the criminality of politicians only matters if the elected winners possess specific characteristics, such as a particular education level or higher amounts of financial resources. Alternative empirical specifications thus include interaction effects for the criminality indicator with either age, wealth, or education to control for potential dependence between the criminality dummy and another control variable to have significant effects on the number of gendered crimes in our data. The adjusted R-squared values of these specifications tell us that the interaction terms mostly slightly improve the model fit compared to the models in which they are not included. Between the new specifications, using the adjusted R-squared as a means to determine whether one interaction effect performs better than another allows us to conclude that the specification including only the interaction between the criminality and the education variables mostly outperforms the other models, except for kidnapping and cruelty. The statistical significance, as well as the magnitude of the state estimates do nevertheless not notably change with respect to the estimation results in Table 3, underlining again the importance of the state characteristics in explaining the prevalence of violence towards women. From the regression output in Table B.3. in the appendix, one can see that the criminality dummy and its interaction terms are, for all specifications, not individually significant.
and insult occurrences at the five percent level. The Wald tests finally show that the other interaction terms are not relevant in improving the specifications. The reason for finding significance of the impact of education coupled with politicians being criminally accused might be that more educated politicians rather resort to criminal acts that involve rent-seeking activities and that are related to their political occupation than to offences against individuals, and more specifically against women. Thus, the reduction of dowry deaths and insults for several levels of education higher than only being literate might be a result of more educated leaders committing different kinds of crimes than less educated political representatives. These leaders possibly resorting to fewer offences against the human body, their behaviour is then reflected by the acts of people that live in the constituencies where these MLAs won the elections. This may hint at better education of elected politicians, suggesting higher quality and competence, being a driver for better outcomes in terms of more secure environments for women and fewer gendered crime incidents, at least for certain types of violence.
5
Discussion
comparable utilising the information included in our data due to having too many between-state dissimilarities. Further, the major states included in this analysis in all likelihood display different base-levels of crimes, making it difficult to construe the regression estimates since the number of committed crimes has a peculiar interpretation for each state: there might be a higher number of rape incidents in one state than in another but the prevalence of incidents could have been fairly declining over the years in the first one, which would be a positive result for that state since this would signify an improvement in the condition of women. Yet, in comparison with the second state, the first one would stand in a bad light by directly comparing them without taking into account the unobserved state-specific heterogeneity in the data.
Since we have retrieved the number of gendered violence incidents from one particular year for each state only and the states seem to be heterogeneous in their inherent characteristics, we are thus not able to draw sound inferences from the estimates. Yet, this caveat could be unravelled by including data on several MLA elections, which take place every five years, and matching each of these elections with data on violence towards women, thereby creating repeated observations on these same states. In that way, it would be possible to draw valid conclusions due to having information on within-state variations and to being able to include year-fixed effects as well. This would allow us to more accurately interpret the variations in crime occurrences since there would be indications on the gender-based crime dynamics and on variations in electing criminal winners over the years, as well as on state-specific fixed effects like culture and religion which have an important role in explaining criminal dispositions. Several studies show that there indeed is a relationship between politicians’ identity (Bhalotra et al., 2012; Iyer et al., 2012) and thus to a large extent their criminality (Chemin, 2012) and the crimes committed in the regions where these politicians are elected. Including within-states observations and using state-fixed effects to control for time-invariant state characteristics in our analysis might thus allow us to properly draw on variations in gendered violence occurrences across districts and constituencies within these states and to only have intra-electoral discrepancies across constituencies.
