What a Crime Statistic Cannot Tell You

A court can count what reaches it. That is not the same as counting what happened. Photo: Raki_Man, via Wikimedia Commons (CC BY 3.0).¹

Tamil Nadu recorded 1,144 cases of crimes and atrocities against Scheduled Castes in 2019. In 2023, it recorded 1,921. The increase was 67.9 per cent.²

That is a serious number. It is also an unusually easy number to misuse.

One political reading says the increase proves that caste violence grew under the DMK. A defensive reading says it proves that victims became more willing to report and police became more willing to register cases. Both explanations are possible. Neither is contained in the series.

The statistic records registered cases. It does not directly record the underlying incidence of caste violence. Between an offence and a row in a government table sit a victim’s willingness and ability to complain, access to a police station, the police response, the legal sections applied, and the system’s rules for counting. A change at any one of those stages can move the number.

The honest conclusion is therefore less satisfying than the partisan ones: the rise is real, but its cause is unidentified.

One number, several moving parts

Suppose the violence in a district stays constant while a new legal-aid network helps more victims file complaints. Registered cases rise. That is an administrative improvement revealing an unchanged harm.

Suppose reporting stays constant but violence rises. Registered cases rise. That is social deterioration.

Suppose both change at once—more violence and better registration. The same number rises faster, but the table still cannot separate the two effects.

It can become more complicated. Publicity around a notorious case may reduce the isolation of later complainants. A police directive may make officers less able to refuse an FIR. Greater Dalit mobility and political assertion may provoke backlash in some places. Legal classification may change even when the incident count does not. The National Campaign on Dalit Human Rights, whose 2026 analysis drew attention to the 67.9 per cent increase, explicitly cautions that its data cover registered cases and are limited by under-reporting and inconsistencies in official records.³

This does not make the statistic useless. It tells us that the formal system encountered far more cases in 2023 than in 2019. What it cannot do is identify which route brought them there.

The dates do not fit a one-party story

The period itself should stop any quick partisan attribution. The years 2019 and 2020 fell under an AIADMK government, 2021 straddled the transfer of power, and 2022–23 fell under a DMK government. The figures rose from 1,144 in 2019 to 1,273 in 2020, 1,376 in 2021, 1,761 in 2022 and 1,921 in 2023.⁴

This is not a controlled comparison between two governments. It is a short time series spanning a change of government, a pandemic, different reporting environments and unknown changes in underlying incidence. Drawing a causal line from the party in office to the final percentage would require evidence the series does not supply.

The opposite claim—this is merely better reporting—has the same defect. It treats a benign explanation as established because it is politically convenient. Better registration may account for some of the rise. So may greater violence. Without independent victimisation surveys, audit studies of police registration, district-level reporting measures and consistent offence-level data, we do not know the proportions.

There is a useful discipline here: when a measure is produced by an institution, first ask what institutional process it measures. An FIR statistic is partly a social-harm measure and partly a police-access measure. It should never be read as if it were a direct sensor placed over society.

Follow the cases downstream

If the headline count cannot settle the causal argument, the justice pipeline tells us something more operational.

Tamil Nadu Police’s 2023 court-disposal table reports 945 completed trials in crimes and atrocities against Scheduled Castes. Of those, 115 ended in conviction and 830 in acquittal. Using the state’s own formula—convictions divided by completed trials—the conviction rate was 12.2 per cent. Of 7,912 cases available for trial during the year, 6,946 remained pending at year-end: 87.8 per cent.⁵

These are not conversion rates for the 1,921 cases registered in 2023. Court cases completed that year can originate in earlier years; annual columns are not a single cohort moving through a funnel. This distinction matters because casually dividing convictions by new registrations would produce a meaningless number.

Even read correctly, a low conviction rate is not proof that every acquittal represents a failed victim. Acquittals can reflect weak investigation, hostile witnesses, intimidation, delay, evidentiary difficulty, misclassification, or allegations that could not be proved. A table cannot allocate the causes.

But a system in which nearly seven out of eight trial-ready cases remain pending, and fewer than one in eight completed trials ends in conviction, has an obvious capacity and quality problem worth investigating. This conclusion does not require pretending to know why each case failed. The backlog is itself a fact about state performance.

The Union government’s own parliamentary reporting shows why the distinction matters. It separately records cases registered, cases charge-sheeted, persons arrested, persons charge-sheeted and persons convicted.² Those are outputs from different stages. They become useful only when analysts resist compressing them into one moral score.

What better public data would look like

The state need not choose between celebrating higher registration and acknowledging persistent violence. It should publish the information that would let citizens distinguish them.

A useful district-level dashboard would report:

• complaints received, FIRs registered and complaints refused or redirected;
• time from complaint to FIR and from FIR to charge sheet;
• cases closed as false, mistaken or civil disputes, with reasons and review outcomes;
• witness-protection requests and action taken;
• completed trials by filing year, not only by disposal year;
• acquittal reasons coded from judgments;
• vacancies and caseloads in police units, prosecution offices and special courts;
• victim surveys capable of estimating harm that never enters police records.

That would not eliminate disagreement. It would make some disagreements answerable.

Tamil Nadu now has a TVK government, with C. Joseph Vijay sworn in as chief minister on 10 May 2026.⁶ It inherits the accumulated case stock; it does not inherit retrospective authorship of every case in it. The fair test is prospective: does reporting become easier without investigation becoming sloppier? Do charge sheets arrive faster? Does pendency fall by filing cohort? Do witnesses remain in cases? Do judgments reveal better evidence?

Those questions are harder to fit on a poster than “cases rose 68 per cent.” They are also closer to justice.

The political rule

When a recorded harm rises, do not automatically congratulate the recording system or condemn the government. First separate three possibilities: the underlying harm changed, the visibility of the harm changed, or both changed.

Then follow the record through the institution. Registration tells us that a case entered. Investigation tells us whether the state built it. Trial tells us whether the evidence survived. Pendency tells us how long the system made everyone wait.

A statistic can expose a problem without explaining it. Respecting that boundary is not evasiveness. It is the beginning of finding out which problem we actually have.

Footnotes

1. https://commons.wikimedia.org/wiki/File:Madras_High_Court_-_panoramio.jpg ↩

2. https://www.mha.gov.in/MHA1/Par2017/pdfs/par2025-pdfs/RS11122025/1183.pdf ↩ ↩²

3. https://www.ncdhr.org.in/wp-content/uploads/2026/02/Five-Years-of-Caste-Based-Atrocity-Feb-9-2026-for-web.pdf ↩

4. https://www.mha.gov.in/MHA1/Par2017/pdfs/par2024-pdfs/RS24072024/301.pdf ↩

5. https://eservices.tnpolice.gov.in/content/crime_review/tn_cr_statistics_2023.pdf ↩

6. https://www.pib.gov.in/PressReleasePage.aspx?PRID=2259495&lang=1&reg=1 ↩