Stabbed in the back? Mandated political representation and murders

Members of the Scheduled Castes and Tribes (SCs and STs) form the social groups who suffer most from caste-based discrimination. This paper focuses on SCs only, leaving aside STs, for two main reasons. First, SCs represent a bigger minority, forming 16 per cent of the Indian population in the 2001 Census as against 8 per cent for STs. Second, the SCs have more frequent interaction with the majority: historically STs live in isolated autonomous villages while SCs belong to multi-caste villages. Today, SCs still represent a significant minority in 80% of rural Indian villages in the nationally representative IHDS 2011, while STs are a significant minority in only 30% of the villages (a significant minority means that the group represents 1–50% of a village population). I denote all non-SC/ST castes as ‘higher caste’.

These crimes may specifically target members of SCs who act as if empowered. For press reports on murders, the most extreme form of caste-based violence in many contexts, see ‘Caste hatred in India—what it looks like’, available from: https://​www.​bbc.​com/​news/​world-asia-india-43972841#. See also Mathew (2003), Narula (1999), Purohit et al. (2002), or Sumathi and Sudarsen (2005) for accounts of caste murders in the political sphere, to prevent lower castes members from running for election to reserved seats, or, when elected, from taking an active part in politics. For example, Mathew (2003, p.156), notes, ‘There has been a sharp increase in violent manifestations of casteism in local communities ever since the local government system got strengthened through the Constitution amendments. Once the panchayati raj institutions were perceived by the upper castes as the tool for the lower castes to assert their rights as individuals living in a democratic polity the latter have become targets of caste-based discrimination and violence’. Such reports contrast with the easy implementation of other forms of affirmative action for SCs, which have an older legacy and a still debated effect (Deshpande 2019; Jaffrelot 2002).

Sabotage is distinct from retaliation in that sabotage is strategic (Brown and Chowdhury 2017), while retaliation is spiteful (Fallucchi and Quercia 2018).

If the coefficient was precisely estimated, members of the SC would declare that they trusted the police less in the presence of SC quotas (p value at 13%). If trust in the police declines, it becomes harder to report crimes when SC quotas are in force, contradicting the interpretation that the quotas bring empowerment. Yet, there is ample room for improvement in the relationship between the police and members of lower castes: in 2001, lower caste members were not allowed to enter police stations in 28% of the villages (Shah et al. 2006).

In Iyer et al. (2012), the increase in crimes against women recorded in India after gender quotas is consistent with the better access of women to justice. Indeed, the number of crimes that are prone to reporting bias (such as rape and kidnapping) increase after gender quotas, but the murders of women do not. Iyer et al. (2012) also report that caste quotas do not affect the number of rapes against members of lower castes. The latter results are puzzling, since SC women are frequent victims of crime and are vulnerable to both gender-based and caste-based electoral quotas. But if both gender and caste quotas had an empowerment effect, shouldn’t SC women have been the ones with the strongest response? Intrigued by this puzzle, I focused on crimes targeting members of SCs and extended the study sample by six states to cover all of India’s 17 major states, and by 6 years to 2013. Taking into account this complete state sample leaves the identification strategy unchanged, but the number of murders of members of SCs then appears to increase after the SC quotas. This result is essential: a precise estimation of the relationship between imposing caste quotas and caste murders is consistent with fundamentally different effects of gender quotas and caste quotas on crimes.

Focusing on SC quotas in the state assembly of Uttar Pradesh, Jensenius (2017) concludes “although SC politicians still experience micro-aggressions, they are spared overt discrimination” (p. 162). The present paper asks what the prospects are for the whole population during quotas in Panchayats—the political institutions closest to the population.

In practice, the IHDS asks only the members of the SCs if they are victims of untouchability practices. The question of practicing untouchability is put to members of all castes. The reported statistics correspond to the responses from non-SCST members, to reflect the analysis of crime data.

The discrepancy between the intensification of crimes against members of lower castes and the population share of these castes is evidence that some crimes specifically target members of lower castes, leading Sharma (2015) to refer to these crimes as ‘hate crimes’.

All data coming from the Censuses of India for 1991, 2001 and 2011 are interpolated to annual values. Exceptions are the crime and police data provided by the NCRB, the real per capita GDP provided by the Ministry of Statistics and Programme Implementation and the electoral data provided by the States Election Commissions, Ministry of Panchayati Raj or Ministry of Rural Development. Urbanization is the share of the state population living in towns.

All the data from the HDPI 1993, the IHDS 2005 and the IHDS 2012 (NCAER 1994; Desai and Vanneman 2005; Desai and Vanneman 2015), are interpolated to annual values. The probability of encounters between members of different castes around a water source transposes the measure used in Bros and Couttenier (2015), with a focus on members of SCs. It is the product of 4 shares: SCs in a state, SC households which have no drinking water on their premises, non-SCST households and non-SCST households which have no drinking water on their premises. Tube wells and taps are excluded from the definition of shared water sources, since they limit the risk of ritual pollution.

The states included are the large states of India, which account for more than 90% of the crimes committed in each of the crime categories. These states are Andhra Pradesh, Assam, Bihar, Gujarat, Haryana, Himachal Pradesh, Jammu and Kashmir, Karnataka, Kerala, Madhya Pradesh, Maharashtra, Odisha, Punjab, Rajasthan, Tamil Nadu, Uttar Pradesh, and West Bengal. Like Iyer et al. (2012), I use the 1992 definition of the frontiers of the states throughout the analysis of state data, to maintain comparable units over time. Hence, I attribute crimes committed within the three new states created between 2001 and 2012 to the states that they initially belong to: I reunite the data of Andhra Pradesh with that of Telangana, of Chhattisgarh and Madhya Pradesh, of Jharkhand and Bihar, and of Uttarakhand and Uttar Pradesh. In two of these cases, the panchayat elections in the new states took place a few years after the panchayat election taking place in the mother state, which introduces measurement error in my estimates. For a visual overview, the data in Fig. 1 cover the exact area that forms the sample of my study (although the data in Fig. 1 span 22 states, because they appear with the current political frontiers which also include the 2019 separation of Jammu and Kashmir and Ladakh).

Any pure effect of SC quotas on reports of special crimes is likely to be difficult to pick up in Table 3 because the law became more stringent on special crimes in 1989, immediately before the start of the crime data sample. Thus, the SC quota may have magnified the impact of the law change in empowering the SCs but the pure quota effect is hard to disentangle from any of the other sources of empowerment trend following the law change. Moreover, the change in the special crime definition may also have led to a reallocation of crimes: offences that were initially recorded under different headings of the penal code became recorded under the heading of special crimes, which may explain why the total number of penal code crimes barely changes with political quotas. In particular, some rapes may have been reallocated to items 11 and 12 of the special crimes (listed in Appendix 13). Murders are the only crimes with such a clear definition that the recording of them was not affected by the 1989 re-definition of special crimes.

These additional controls have the expected relationship to crimes. The share of votes for the BSP has a negative relationship, if any, with crimes. Interestingly, if I interact this vote share with the post-reservation dummy, the interaction term has a significantly negative relationship with murders, attenuating the negative main effect of the post reservation dummy. Consistent with Sharma’s findings on district data from 2001 to 2010, an increase in the income of non SCST households translates into a significant decrease in overall and penal code crimes (Sharma 2015). Consistent with the findings of Bros and Couttenier (2015) on district data from 2001 to 2011, an increase in the probability of encounter has a positive but imprecisely estimated relationship with the number of murders (and with the number of rapes).

Critical values are drawn from a t-distribution to account for the small number of clusters, where the degree of freedom is equal to the number of clusters minus the number of regressors that do not vary within the clusters. I use 1000 replications.

The framework allows us to formalize why one individual may choose an action that seems economically absurd at first sight but makes sense if the identity group norms are considered and that these norms craft the individual payoff. For example, Akerlof and Kranton (2010) demonstrate that if there is a norm in a black identity group to not work productively for members of a white identity group, the members of the black identity group will lose some economic payoff (the wage that they forgo) but will gain some identity payoff (through their peers or themselves for doing what is deemed right in their value system). In the case of a murder motivated by an identity concern, the perpetrator may lose her freedom of movement (if she ends up going to jail), but gain some identity payoff (peer recognition or a warm glow).

Narula (1999), for example, reports that: “In the village of Melavalavu, Madurai district in Tamil Nadu, following the election of a Dalit to the village council presidency, members of a higher-caste group murdered six Dalits in June 1997, including the elected council president, whom they beheaded”. See Footnote 4 for more examples.

I assess the consequence of excluding or including different states in Appendix Table 8. In column 1, I restrict the sample to the 11 states used by Iyer et al. (2012) to study caste quotas. In column 2, I restrict the sample to the 13 states which implemented SC quotas in 1993 or later, so that none of the states is affected by quotas during the first year for which crime data are available. In column 3, I use all 17 states. The magnitude of the coefficient changes importantly between columns 1 and 2, but not between columns 2 and 3. Column 3 estimates are more precise than column 2 estimates, but the magnitudes of the coefficients are barely affected. Such observations lend support to working with the full sample of 17 states. Note that on the samples with all 17 states, since four states implemented SC quotas in 1992 or before, and since the SC crime data start in 1992, these four states appear as “always treated” (however, since all the other states implemented quotas after 1992, the quota variable is not a linear function of the states’ fixed effects). Importantly, including these four states neither affects the identification strategy, nor leads to statistically different magnitudes of estimates, but it does reduce standard errors. Last, the extension of the sample by six years does not lead to any significant changes (see column 4 of Appendix Table 8).

The IHDS data do not allow us to keep track of cases when SC and gender quotas overlap: these cases may have been recorded either in the SC quota, or in the gender quota treatment, introducing measurement error in my estimate. However, such error would introduce an attenuation bias since it means that the control group may be contaminated by the treatment.

Table 4 shows the absence of correlation of victims’ and perpetrators’ answers in the IHDS raw data. Duncan 1976, Baumeister et al. 1990 discuss this divergence in other contexts. Krumpal (2013) underlines the risk of social desirability bias in survey responses, a bias that is likely to affect the claims by victims and by perpetrators differently. However, even if the survey questions on untouchability are not the ideal measures of absolute levels of discrimination, the magnitudes of their reaction to quotas may tell us something about the impact of the quotas (Bertrand and Duflo 2016).

Although most elections in the sample take place during the post quota period some states did run elections before quotas were implemented. These results are robust to alternative definitions of the electoral period (tables available upon request).

A virtuous interaction could ensue if, for example, the special courts were to function better after quotas had been implemented, or if the special courts could top up the empowerment effect of political quotas. Otherwise, a vicious interaction might arise if, for example, members of higher castes were more likely to practice sabotage in places where inter-caste relationships were becoming tenser. So far, nine of the sample states (namely Andhra Pradesh, Bihar, Gujarat, Himanchal Pradesh, Karnataka, Madhya Pradesh, Rajasthan, Tamil Nadu and Uttar Pradesh) have an exclusive special court in one district at least. I create a binary variable that takes the value one for states with exclusive special courts because the date of the actual implementation of exclusive special courts is not, to the best of my knowledge, public knowledge in every state. This crude coding induces an attenuation bias for empirical results, and it does not allow me to compute the main effect of an exclusive special court (this is eliminated by state fixed effects).

An amendment to the Prevention of Atrocities Act passed in 2015 aims, among other things, to improve the quality of service provided by the special courts. Its efficiency in facilitating the access to justice for the SCs is to be evaluated in a future work.

The 73rd amendment of the Constitution requires SC quotas to mirror the proportion of SCs in each state population. I use the share of SC in each state population at the last census before an election to measure the proportion of Panchayat seats reserved for members of the SCs. Since there is one census per decade, but the population share of SCs changes continuously, by controlling for the interpolated population share, I can identify the effect of increased SC representation due to the within-state variation resulting from the adjustment in quota size at each election following a census. This strategy is similar in essence to that of Pande (2003) for the size of SC quotas in state elections.

This rate of election of SC candidates is significantly smaller than the SC household share in these villages which is 23%.