Research insight
March 12, 2021

Atrocities, murders, and untouchability: Measuring caste-based discrimination

In order to combat caste-based discrimination, the Indian State implemented the Prevention of Atrocities Act, 1989 (henceforth, PoA Act). The Act focusses on offences related to the persistence of caste-based practices that are outlawed today, and the intentional humiliation of members of the lower castes. The PoA Act provides stronger punishment than what would be the penal code ruling, for a range of issues that are symbolically sensitive. For example, if a higher caste member denies access to a water source to a member of a Scheduled Caste/Tribe (SC/ST).1

As a result, an increase in the number of cases registered under PoA Act is often interpreted as a negative signal of caste-based discrimination and tensions. For example, the Union minister of state for social justice and empowerment recently cautioned Telangana government on the high number of cases booked under the Scheduled Caste and Tribes (Prevention of Atrocities) Act, 1989 after the state formation, and that the state has a much bigger number of cases compared to many bigger states.

However, as PoA Act cases depend on interactions between at least three actors, heightened tensions may not always translate in higher cases. The recording of a PoA Act case depends on not only the perpetration of the atrocity, but also the willingness of the victims to report, and of the police to record this atrocity – as stressed in Iyer et al. (2012) for crimes targeting women. Caste-based discrimination may prevent the recording of cases under PoA Act – for example, discriminatory behaviour by police officers may take the form of actively preventing the recording of offences, or an outburst of tensions may deter victims from trying to report the offences (Deswal 2013, Kumar 2015, Minj 2018).

Comparing findings from different measures of caste-based tension

This post builds upon data in Girard (2020), to contrast three different potential measures of the tensions related to the perpetuation of caste-based hierarchies: (i) caste-based murders, (ii) PoA Act offences, and (iii) survey declarations on untouchability.

I use SC murders as a benchmark of the level of violence targeting members of the SCs. This approach, used in Aneja and Ritadhi (2020), Bros and Couttenier (2015), Girard (2020), and Iyer et al. (2012), is based on the idea that murder is the type of crime where the reporting bias is the lowest, if only because a body is hard to hide.

The visual representation in Figure 1 of the cases, as recorded by the police, casts doubt on the suitability of the use of the PoA Act cases as a measure of caste-based discrimination. PoA Act cases display no clear pattern, not even in states of the Hindi belt.4 In contrast, murder records are the highest in the states of the Hindi belt. Rajasthan, for example, records relatively few PoA Act cases, while it is one of the states witnessing the highest rate of murders per 100,000 members of the SCs.

Figure 1. SC murders and PoA offences, 2012

We can also check the police cases on PoA and murders against households’ survey responses about whether they either practise or have faced untouchability. Responses to these survey questions may be hard to interpret as they straightforwardly ask about a practice that has been anti-constitutional for more than half a century. We may worry that such straightforward questions on touchy subjects induce a social desirability bias in responses and the resulting data may not reflect ground realities well.

Unlike PoA Act cases, household survey responses on untouchability, when averaged by state, document a pattern strikingly consistent with that of caste-based murders. Figure 2 shows that untouchability practices are particularly widespread in the Hindi belt and are lower elsewhere. The Figure reveals a visual correlation between the share of non-SC/ST households that have declared that they practice untouchability and the share of SC households declaring that they suffer from untouchability.

Figure 2. Untouchability practice and victims across states

Note: The Figure shows percentage of non-SC/ST households who reported practicing untouchability, and percentage of SC households that reported being victims of untouchability

On top of concerns with respect to social desirability bias in household responses, murders are a particularly extreme way of enforcing caste-based discrimination, and much more extreme than the daily practice of untouchability. We may thus expect the two behaviors to follow fundamentally different patterns. Yet, the murders and household survey measures appear to display a strong cross-sectional correlation.

For a statistical test of the visual correlations in the Figures 1 and 2, Table 1 displays the Spearman rank correlation matrix.5 Murder cases are positively correlated with households’ declaration on either perpetrating untouchability or suffering from untouchability practices. On the contrary, the PoA Act cases are not significantly correlated with either murder cases or households’ declarations.

Table 1. Correlation between police cases and householdsurvey responses


(i) The table displays the Spearman rank correlation matrix.

(ii) At the state level, the number of murders per 100,000 SC members in 2012 displays a correlation of 48.4 with the share of members of the SC who declare having faced untouchability practices, and this correlation is statistically significant at the standard 5% threshold. A significance level of 5% indicates a 5% risk of concluding that the strength of the relationship happened by chance if the null hypothesis were true.

(iii) The correlation between practice of untouchability and facing untouchability is significant at the 1% threshold.

Key takeaways

The set of comparisons above indicates three important facts:

(i) Murder cases and survey declarations consistently reveal variations in inter-caste tensions across states, which the PoA Act cases do not.

(ii) Statistics from offences recorded after the PoA Act should be interpreted with a lot of caution. To take an extreme illustration, these cases could be at similar levels in a place where there is almost no reporting to the police of highly prevalent discriminatory practices, and in another place where there is systematic reporting of the few instances of discriminatory practices.

(iii) State-level averages of two survey measures of untouchability practice and victimhood sheds new light on the prevalence of untouchability practices.

This blog note was initially published in voxdev/ideas4india

Victoire Girard

Victoire Girard is a researcher at NOVAFRICA, Nova SBE. She is also a research affiliate of the LEO, Orléans University, and an invited lecturer at Paris 1 Panthéon-Sorbonne University where she defended her PhD.

1. No poverty
10. Reduced inequalities
Research insight

Are Muslim immigrants really different? Experimental Evidence from Lebanon and Australia

Under the theme of Economic Development in Africa, the NOVAFRICA@Nova SBE seminar series hosts a variety of international speakers. This seminar invited Maleke Fourati, University of Geneva.

Learn more
Opinion Article

Economics at the Service of People

The NOVAFRICA’s vision of the Nobel Prize in Economics in 2011. A deserved distinction for tackling the most important challenges in fighting poverty. NOVAFRICA celebrates being part of this movement since 2011.

Learn more
Opinion Article

Trust, complementarity, and commitment

In a few years, all companies that survive will be more innovative, more agile, more focused on their purpose, more sustainable. In a word, it will be a new organizational culture for the future.

Learn more