Crime statistics are among the most political numbers a state produces, and among the least self-explanatory. When recorded hate crime rises in Britain or the United States, officials and activists routinely dispute whether the streets have grown more dangerous or the forms easier to file, and both sides are usually partly right. India now faces the same interpretive problem, at much larger scale, with the numbers that track violence against its most marginalized citizens.
The occasion is a five-year review published this year by the National Campaign on Dalit Human Rights (NCDHR), an advocacy coalition, analyzing official data on crimes against Scheduled Castes and Scheduled Tribes: the constitutional categories covering Dalits, the communities once branded “untouchable,” and Adivasis, India’s indigenous peoples. Drawing on the National Crime Records Bureau’s own publications, the report counts 45,935 recorded caste-based crimes against Scheduled Castes in 2019, rising to 57,789 by 2023, an increase of roughly 26 per cent, and a five-year total of 262,474 cases. Crimes against Scheduled Tribes rose faster still over the period. State-level findings sharpen the picture: in Tamil Nadu, a state that has long projected itself as a model of social-justice governance, the coalition’s analysis found a 67 per cent jump in atrocities against Dalits and Adivasis over the same five years, a finding that landed uncomfortably in a state election year. The pattern isn’t news to people who study it. Speaking at a Dalit Intellectual Collective consultation on caste violence in the state in January 2024, Professor Sukumar of Delhi University’s political science department put the same contradiction more bluntly than any statistic can: Tamil Nadu should “either do something to sensitise caste Hindus to create a balanced society or stop calling this place a land of social justice.”
The report’s framing treats the climb as evidence of rising atrocity. It may well be. But a recorded-crime series cannot, by itself, establish that. A rise in registered cases is equally consistent with a rise in registration: more assertive complainants, more Dalit awareness of the Prevention of Atrocities Act, marginally less obstruction at the police-station counter. Criminologists everywhere treat growth in reported hate crime as an ambiguous signal for exactly this reason, and India adds a further wrinkle: under-registration of caste crimes has historically been so severe that better recording would itself be a form of progress. A dataset built entirely from official statistics has no instrument for separating more violence from more visibility. Honest analysis has to hold both possibilities open.
But the argument should not stop there, because the same official data speaks far less ambiguously about what happens after a case is registered, and that is where the report’s indictment lies. The Scheduled Castes and Scheduled Tribes (Prevention of Atrocities) Act of 1989 did not merely criminalize caste violence; it built special machinery (designated courts, special prosecutors, mandated timelines) precisely because the ordinary system was too slow and too porous for these cases. The government’s own figures show what that machinery has become: NCRB data compiled in NCDHR’s own five-year review put 332,606 cases pending before the courts in 2023 alone, of which 311,986 remained unresolved at year’s end — a 93.8 percent pendency rate. Pendency on that scale means the special courts now hold a backlog larger than the last five years of new registrations combined. Earlier data points the same direction: of the roughly 50,000 SC-crime cases in 2020, investigation was still pending in nearly 20,000 at year’s end. The bottlenecks the new report names (judicial delay, inconsistent police follow-through, social pressure on complainants to withdraw) are the same bottlenecks named by reports on the same law for a decade.
Source: NCDHR five-year review of NCRB data, 2023. Compiled by The Dalit Project.
Read together, the two halves of the data point to a specific diagnosis. Whatever the true trend in violence, the law’s binding constraint is not the statute and not, primarily, deterrence. A special court that takes years to reach trial does not deter; a complainant facing village-level pressure to withdraw, with no protection during the wait, converts a registered case into a withdrawn one; a police force that files the FIR but slow-walks the charge sheet produces exactly the pattern the numbers show: rising registration flowing into a static, clogged pipeline. The chokepoint was never the paper right. It is everything downstream of the complaint.
That distinction matters because it forks the policy response. One family of remedies (harsher sentencing, awareness campaigns, ritual condemnation) treats the problem as attitudes that have not yet absorbed the law’s existence. The other (special-court staffing and dedicated dockets, witness and complainant protection against withdrawal pressure, accountability for investigative delay, published timelines per district) treats it as a system that already has the right law and the wrong throughput. The data in this report is far more consistent with the second problem than the first. Whether the political system prefers the first anyway is a separate question; deterrence-flavored responses are cheaper, more announceable, and do not require confronting police forces and court administrations about their own performance.
There is a caution owed in both directions here. Advocacy groups have understandable reasons to present rising numbers as rising atrocity (ambiguity does not mobilize) but the strongest version of their case does not depend on that reading. If violence is rising, a 300,000-case backlog is a catastrophe; if reporting is rising, the backlog is still a catastrophe, because it means the state invited complaints it cannot process. The bottleneck indictment survives either interpretation of the trend line, which is precisely why it, rather than the headline percentage, deserves to lead the argument.
The forces underneath will not wait for the interpretation to be settled. Dalit assertion (legal, digital, electoral) is growing, which alone should push registration upward for years; the court system’s capacity is not growing to match; and each case that dies of delay teaches a village, more durably than any statute, what the law is worth. The next five-year review will almost certainly report a larger number. The question worth asking now is whether it will also report a shorter queue, because the queue, not the crime count, is the figure that will say whether the law works, and nothing in the machinery is yet moving to shorten it.
