The Conviction Gap: What NCRB's Own Data Shows About the Prevention of Atrocities Act

Rajmani Devi waited sixteen years. In January 1999, the Ranvir Sena, an upper-caste militia in Jehanabad district, Bihar, killed at least 23 Dalit villagers in the Shankar Bigha massacre. Rajmani Devi lost her husband, her mother-in-law, her father-in-law, and her sister-in-law that night. The case went to trial. It stayed there for sixteen years. On 13 January 2015, the sessions court acquitted all 24 defendants for lack of evidence. “All these years, I had kept waiting for justice hoping the accused persons will be served harsh punishment; instead they have been acquitted,” she told Gulf News at the time. “I am totally hopeless now.”

Etwaria Devi, who lost her husband and son that same night, told the same reporter something closer to a verdict on the system itself than on the case: “There is no one to give justice to us, sir. We lost our world 16 years back and now all our hopes stand dashed after the court verdict. This is the way a poor person is treated here.” Comrade Mahanand, the CPI-ML’s Jehanabad district secretary, put the arithmetic of it more bluntly: “It is a fact that 23 Dalits were killed in the massacre by the Ranvir Sena men. The court should now say who then killed the poor if all the accused persons are innocent?”

Shankar Bigha is one case. But pull two years of the National Crime Records Bureau’s own published court-disposal tables, and Rajmani Devi’s sixteen years stop looking like an outlier. They look like the system working as designed, on a schedule that hasn’t moved. The chart above states the finding: of the small share of cases India’s courts actually decide, most don’t end in conviction.

Of every 100 Scheduled Caste atrocity cases that reached a verdict in an Indian court in 2022, 34 ended in conviction. Sixty-two ended in acquittal. Four were discharged before trial concluded.

But that 34% describes only the cases that got decided. In 2022, 300,303 Scheduled Caste atrocity cases were sitting before India’s courts waiting for trial. Only 16,538 of them, 5.5%, reached a verdict that year. The conviction rate describes a small, non-random slice of a much larger backlog. The backlog itself is the more brutal number here, not the odds once a case finally gets heard.

This is not a one-year problem

The instinct with a number like 5.5% is to check whether it was an unusually bad year. It wasn’t. NCRB’s 2021 court-disposal tables, extracted directly, show the same pattern. In 2021, 264,694 Scheduled Caste atrocity cases were awaiting trial. Only 10,120 of them, 3.8%, were decided that year, an even smaller share than 2022’s. Of those, 36.0% ended in conviction, a fraction of a point higher than 2022’s 34.0%, and functionally the same story: roughly a third convicted, roughly two-thirds not, on a caseload that grew 13.5% in a single year while the courts’ capacity to decide cases barely moved with it.

Widen the lens further and the same conclusion holds on the registration side too, across four consecutive years now directly extracted: 45,961 SC atrocity cases registered in 2019, 50,291 in 2020, 50,900 in 2021, 57,582 in 2022. Every year, more cases. Every year, a courts system that decides a shrinking share of a growing pile. This is what the NCRB has been showing, release after release, for as long as it’s been verified directly against the primary tables. It is not a finding. It is a standing fact the country’s own government has published annually and moved past annually.

Where the backlog breaks hardest

Not every state carries this backlog the same way, and the three-pattern split that showed up in 2022 shows up again, independently, in the 2021 data.

Bihar carries the second-largest caseload of any state and moves through it slower than almost anywhere else, in both years measured: 0.6% of pending cases reached a verdict in 2022, 0.3% in 2021. Bihar’s number isn’t abstract. It’s Shankar Bigha, and Etwaria Devi’s sixteen years, repeated across a docket of 58,522 cases waiting behind hers.

Karnataka moves through cases faster than Bihar in both years, but of the cases it decides, 98.7% in 2022 and 98.4% in 2021 end in something other than conviction. This is not new. In 2011, a coalition of Karnataka Dalit rights organizations, including the National Campaign for Dalit Human Rights and the Karnataka Dalit Mahila Vedike, published a study finding the state’s own atrocity conviction rate at 5% for 2010, with seven districts recording zero convictions and Tumkur district alone disposing 101 cases without a single one, reported at the time by the Deccan Herald. The same year, P.L. Punia, then chairman of the National Commission for Scheduled Castes, told reporters in Bangalore that Karnataka’s conviction rate stood at roughly 9%, “one of the lowest in the country,” and asked pointedly, “in many States, the conviction rate in nearly 50 per cent. If other States can do it, why not Karnataka?”, per the Deccan Herald’s coverage from the same year. Fifteen years later, direct extraction of NCRB’s 2021 and 2022 data puts Karnataka’s conviction rate at 1.6% and 1.3%. Whatever changed in Karnataka’s atrocity courts between 2011 and now, the conviction rate is not it.

Uttar Pradesh is the outlier in the other direction in both years: the largest caseload, and, among cases it does decide, the highest conviction share of the five. Madhya Pradesh and Rajasthan sit between the extremes: Madhya Pradesh clears its backlog faster than any state in this table, 10.3% decided in 2022, ahead of even Uttar Pradesh, while converting only 22.9% of that into convictions, closer to Karnataka’s pattern than Bihar’s. Rajasthan lands in the middle on both counts, and its steadiness across both years is itself informative, a baseline against which the three sharper patterns stand out. There is no single national story here. There are at least three different failure patterns, stable across both years verified directly here, and a state-level accountability push aimed at “the conviction rate” without knowing which pattern a given state has would be aiming at the wrong lever in most of the country.

Why convictions don’t follow

None of this happens without a mechanism, and the mechanism is documented, though not by any single source and not as one undifferentiated finding. In short: special courts that were never established, prosecutors who aren’t staffed against the courts that do exist, chargesheets filed months past a legal deadline, and oversight committees that meet a fraction as often as the law requires. What follows is the paper trail behind that sentence.

The Union Ministry of Social Justice and Empowerment’s own 2022 annual report under Section 21(4) of the Act found that only 194 of the 498 districts legally required to have one had established the mandatory special court, per reporting in The Wire, which also cites the Ministry’s own figure for the national conviction rate falling from 39.2% in 2020 to 32.4% in 2022. That figure sits a few points below the 34.0% directly extracted here for 2022; the gap is most likely a different denominator or a since-revised NCRB release rather than a factual dispute, and it doesn’t change the direction or the scale of the problem either way.

The gaps compound down the line. A citizens’ audit by the Citizens’ Vigilance and Monitoring Committee, examining the government’s own Section 21(4) reporting, found via The Leaflet a shortage of special public prosecutors relative to the special courts that do exist, no exclusive special courts at all in several high-atrocity states including Haryana and Kerala, and the legally mandated state-level monitoring committees meeting only 12% of the time they were supposed to in 2021. Statutory timelines slip earlier too: the Dalit Human Rights Defenders Network, drawing on Ministry of Social Justice data, found that chargesheets were filed after the Act’s mandatory 60-day limit in 46.8% of cases registered between 2016 and 2020 nationally, with Telangana the worst-performing state at 80% filed late, reported by The News Minute. The National Coalition for Strengthening SC & ST PoA Act names similar process failures directly, shoddy investigation, biased witness statements, late chargesheets, weak victim support, in a statement covered by SabrangIndia, and India’s Parliamentary Standing Committee on Home Affairs reached a similar conclusion in March 2021, attributing low conviction rates to “poor implementation of existing laws and apathetic attitude of law enforcing agencies,” per Oxfam India’s summary of the committee’s findings.

Peer-reviewed research offers a mechanism for some of this, not just another count of the same failure. Anthropologist Sandhya Fuchs’s 2024 study of the Act’s implementation, published in the Journal of the Royal Anthropological Institute, found that police and judges routinely conflate three distinct kinds of “truth” when weighing a caste-atrocity complaint: whether the complainant seems personally credible, whether the evidence clears a narrow technical-legal bar, and whether the discrimination described happened. Collapsing the three into one judgment, Fuchs argues, tends to recast Dalit complainants specifically as unreliable or self-interested, a filter that operates before any of the institutional gaps above ever come into play. A separate, earlier study names an even earlier filter, upstream of anything the court-disposal tables above can show. Examining 100 police “Final Reports,” the closures filed before a case ever gets a chargesheet, sociologist Sthabir Khora found in a 2014 Economic and Political Weekly study that police disproportionately marked SC/ST Act complaints “false” rather than “mistake of fact,” a distinction that determines whether a complainant can then be prosecuted for filing a false case. It’s a small, qualitative sample, not a national rate comparable to NCRB’s own figures, but it names an entry point for the same failure the aggregate data only shows from the far end.

None of these sources agree on a single number. What they agree on, independently, working from different data and different years, is where in the process the failure sits: not at the point of registering a complaint, but in the machinery built to carry a registered case through to a verdict, special courts that don’t exist where they’re required to, prosecutors who aren’t staffed against the courts that do, chargesheets filed months past deadline, and monitoring committees that meet once a year when the law expects them to meet regularly. A system can be doing all of that and still, on paper, be “handling” a rising caseload.

The part police are usually blamed for isn’t where this breaks

The common assumption is that police aren’t registering or investigating these cases seriously. NCRB’s own numbers complicate that story rather than confirming it. Two different numbers exist for “the chargesheeting rate,” and which one gets used changes the picture considerably.

NCRB’s own headline chargesheeting rate for SC cases in 2022 is 79.9%. That number divides charge-sheeted cases by cases police have finished handling, which excludes the roughly 19,000 cases still under active investigation at year’s end. A second, equally legitimate number, dividing charge sheets filed by every case police touched that year including ones still under investigation, comes out closer to 60%. SCC Online’s coverage of this data reports a figure in that range, without specifying which of the two definitions it’s using. Neither number is wrong. They answer different questions, and a reader deserves to know which one they’re looking at. Either way, both numbers say the same thing about where the bottleneck sits: police are filing charge sheets in a clear majority of cases, however you count it. The system doesn’t start failing at the police station. It fails after that, in a court system that decides 5.5% of its backlog in a year and convicts a third of what it does decide, for reasons the reports above have already named.

What this costs

Rajmani Devi’s sixteen-year wait ended in acquittal, the outcome 62% of decided cases reach. But the more precise thing NCRB’s data says about her case is this: for every year it sat in the system, the odds it would be decided at all that year were roughly one in twenty, worse in 2021, when the same odds were closer to one in twenty-six nationally. A backlog this size isn’t a delay that resolves itself with patience. It functions as a second form of the same denial the original violence inflicted, one where the state’s failure to act is diffuse and slow enough, spread across annual data releases anyone could read and few do, that no single official has to answer for it, the way the original perpetrators eventually didn’t either.

Etwaria Devi asked, in effect, who is accountable for that second denial the same way she’d asked who was accountable for the first. NCRB’s own data, extracted directly and checked against itself across two years and four, answers that question by not changing. Next year’s release will very likely say the same thing this year’s did. The test of whether that’s finally treated as a crisis rather than a baseline isn’t whether the number moves. It’s whether anyone in a position to fix a special-court vacancy or a missed sixty-day deadline reads it that way before the next release confirms it again.

The 2022 and 2021 figures above are extracted directly from NCRB’s own published tables, not secondary reporting; other historical and causal claims are drawn from named, cited third-party reporting and government committee findings.