Crime statistics: NCRB data and crime mapping

In short

Crime statistics are the official counts of reported crime, victims, accused and case disposal compiled in India mainly by the National Crime Records Bureau (NCRB), a Ministry of Home Affairs body set up in 1986, from returns sent by every State and Union Territory through its State Crime Records Bureau. Crime mapping takes those records down to the police-station, beat and street level and plots them in a Geographic Information System to find hotspots, patterns and trends that drive patrolling, picketing and resource deployment. The topic matters because policy, budgets, Supreme Court directions and media debate all rest on these numbers, and the numbers have known, examinable limitations - under-reporting, the principal offence rule and population-base problems. It is examined in UGC-CBCS criminology papers (Criminal Statistics / Research Methods in Criminology), in UPSC CSE and State PSC police papers, and asked in interviews for crime analyst, research associate and data roles with police departments, BPR&D, NIPFP-type institutes, NGOs and the private risk-analytics industry.

Explained

Who produces crime data in India, and under what law

• NCRB (New Delhi, under MHA, established 11 March 1986, on the National Police Commission's recommendation, by merging the Directorate of Coordination and Police Computers, the Inter-State Criminals Data Branch and the Central Finger Print Bureau of the CBI, and the statistical section of BPR&D) is the nodal agency. It publishes annually:
  • Crime in India (CII) - the flagship volume on cognizable crime, since 1953.
  • Accidental Deaths and Suicides in India (ADSI) - since 1967.
  • Prison Statistics India (PSI) - prison population, overcrowding, undertrials.
  • Finger Print in India and the Crime and Criminal Tracking Network and Systems (CCTNS) reports.
• Flow of data: police station FIR and final report → District Crime Records Bureau → State Crime Records Bureau (SCRB) → NCRB in prescribed proformae. NCRB does not collect from citizens directly; it is an administrative (police-recorded) data system.
• Legal base of the record itself: an FIR in a cognizable case is registered under section 173 of the Bharatiya Nagarik Suraksha Sanhita (BNSS), 2023 (the former CrPC section 154), which also provides for e-FIR / electronic registration and for zero FIR; the final report (charge-sheet) is filed under BNSS section 193 (former CrPC 173). The substantive crime heads now follow the Bharatiya Nyaya Sanhita (BNS), 2023, which replaced the Indian Penal Code, 1860 from 1 July 2024 - so recent CII volumes report BNS heads while older series report IPC heads, and the two are not one-to-one comparable.
• Other data holders: BPR&D (police organisation data - *Data on Police Organisations*), NHRC (custodial deaths), NCW, Ministry of Women and Child Development, NFHS (self-reported spousal violence), and the courts' National Judicial Data Grid. NATGRID links databases for investigation, not for publication.
• Crime is a State subject (Public Order and Police, State List entries 1 and 2, Seventh Schedule), so the Centre compiles but does not record.

Core concepts, classifications and counting rules

• Cognizable vs non-cognizable: CII counts cognizable offences - those where police may arrest without warrant and investigate without a magistrate's order. Non-cognizable offences (NCR entries) are largely outside the headline tables.
• Crime heads:
  • IPC/BNS crimes - body (murder, hurt, kidnapping), property (theft, robbery, dacoity, burglary), public order (rioting), economic (cheating, forgery, criminal breach of trust), women- and child-specific heads.
  • SLL - Special and Local Laws - NDPS Act, Arms Act, Excise, Gambling, POCSO, IT Act, Immoral Traffic (Prevention) Act, SC/ST (Prevention of Atrocities) Act, PCPNDT, Prohibition of Child Marriage Act, etc.
  • Total cognizable crime = IPC/BNS + SLL. Many comparisons go wrong because SLL registration is a matter of enforcement drive, not of crime incidence (a liquor-prohibition State records huge Excise numbers).
• Principal Offence Rule (POR): where one incident involves several offences, NCRB counts only the most heinous one (by maximum punishment). A dacoity with murder is counted as murder. Consequence: lesser offences are undercounted and incident counts are smaller than offence counts. NCRB has, in recent volumes, additionally published some data without strict POR for women-related crimes - the direction of travel is towards multiple-offence tabulation, but POR remains the classical examined answer.
• Unit of count: CII counts cases (incidents) registered, not victims, except in separate victim tables. One murder with three victims is one case, three victims.
• Case disposal terms:
  • "True" cases vs cases declared false or mistake of fact/law.
  • Charge-sheeting rate - charge-sheets filed as a share of cases investigated (police performance).
  • Conviction rate - convictions as a share of cases in which trial was completed (court outcome).
  • Pendency percentage - cases pending investigation or trial at year-end.
• Crime clock: a presentational device - "one crime every X seconds" - derived by dividing the period by the number of crimes; it has no analytical value.
• Juvenile data uses "children in conflict with law" under the Juvenile Justice (Care and Protection of Children) Act, 2015; "juvenile crime" and "juvenile delinquency" are not interchangeable with adult crime counts.

Measures: rate, ratio, trend, and the dark figure

• Crime rate (per 100,000 population) is the comparable measure; absolute numbers favour large States (UP and Maharashtra always "top the list" on volume).
• Crime rate uses projected population, not the 2011 Census count. Because the decennial Census due in 2021 was deferred - the next Census is being conducted with reference date 1 March 2027, and 1 October 2026 for the snow-bound areas - every recent crime rate rests on projections by the Technical Group on Population Projections. Rates for the most recent years should be read as provisional on that account.
• Incidence vs prevalence: incidence = new cases in a period; prevalence = proportion of the population victimised at least once (needs a survey, not police records).
• Dark figure of crime / crime funnel: crimes committed → perceived → reported → recorded → charge-sheeted → convicted. Each stage sheds cases, so recorded crime is a lower bound.
  • Reasons for non-reporting: fear of reprisal, stigma (sexual offences), distrust of police, cost and time, triviality, no insurance need, family settlement, police refusal to register ("burking").
  • Measures of the dark figure: victimisation surveys (India has no regular national one; the Common Cause and Lokniti-CSDS *Status of Policing in India Reports* and the Delhi victimisation surveys are the nearest), self-report studies, hospital and insurance records.
• Recording bias both ways: better policing, mandatory registration after *Lalita Kumari v. Govt. of U.P.* (2014, Supreme Court - FIR registration is mandatory in cognizable cases), helplines and e-FIR all raise recorded crime without raising actual crime. A fall in recorded crime may mean suppression.
• Comparability breaks at: IPC→BNS transition (1 July 2024), new heads added (cyber heads, POCSO after 2012, SC/ST Amendment 2015), the 2014 switch in some women-related tabulations, and changed definitions (for example, rape after the Criminal Law (Amendment) Act, 2013).

Crime mapping: concepts and methods

• Definition: the systematic use of spatial data and GIS to visualise, query and analyse where crime occurs, in order to explain and prevent it. Historic roots: the cartographic school of criminology - Guerry and Quetelet (1830s France, "moral statistics"), later the Chicago School's concentric-zone mapping (Shaw and McKay, social disorganisation).
• Theory that justifies mapping: routine activity theory (Cohen and Felson - motivated offender + suitable target + absence of capable guardian converge in space and time); rational choice; crime pattern theory (Brantingham - nodes, paths, edges, awareness space); broken windows; CPTED (crime prevention through environmental design).
• Geocoding: converting an address or landmark in the FIR into coordinates (latitude/longitude). Accuracy is the weak link in India - vague addresses, no uniform street addressing, "near X temple" descriptions.
• Mapping types:
  1. Pin / point map - one symbol per incident; simple, but overlaps and cannot show density.
  2. Graduated / proportional symbol map - symbol size by count per beat or station.
  3. Choropleth (thematic) map - areas shaded by rate; needs rates, not counts, else it just shows population.
  4. Grid / fishnet map - uniform cells, avoids unequal administrative units.
  5. Density map (kernel density estimation, KDE) - a smooth surface of crime intensity; the standard hotspot output.
  6. Spatio-temporal / aoristic analysis - crime by hour, day, month; aoristic analysis spreads an incident over its possible time window when the exact time is unknown (common in burglary).
  7. Risk terrain modelling - maps the risk from environmental features (liquor vends, bus stands, ATMs, unlit stretches) rather than past points alone.
  8. Network analysis - crime along road segments rather than areas.
• Hotspot statistics: nearest-neighbour index, Moran's I (global spatial autocorrelation), Getis-Ord Gi\* (local hotspot/coldspot z-scores), Ripley's K, standard deviational ellipse, DBSCAN/STAC clustering. A "hotspot" is a statistically significant cluster, not merely a dark patch on a map.
• Repeat and near-repeat victimisation: the same target, or a target close by, is at raised risk shortly after an offence - the empirical basis of hotspot patrolling and of "predictive policing".
• Operational uses: beat redesign, night picketing, Nirbhaya-fund dark-spot and women-safety audits, traffic black-spot identification (road accident mapping under MoRTH), narcotics source mapping, missing-person and vehicle-theft corridors.
• Indian systems: CCTNS (Crime and Criminal Tracking Network and Systems, a Mission Mode Project under the NeGP, covering the large majority of police stations) is the data spine; ICJS (Inter-operable Criminal Justice System) links police, courts, prisons, prosecution and forensics; State dashboards and apps such as Delhi Police's mapping cells and various State "crime GIS" projects consume it. Software used: QGIS (free), ArcGIS, Google Earth, GeoServer, Python (GeoPandas), R (spatstat), plus the US-origin CrimeStat and the classic ArcGIS Hot Spot Analysis tool.

Limitations, ethics and law

• Statistical limitations: POR undercount; SLL driven by enforcement; projected denominators; no national victimisation survey; State-to-State definitional drift; year of registration not year of occurrence; "crime rate" for a crime against women computed on total population versus on female population gives very different numbers (NCRB uses the relevant population group for some heads - state which base you use).
• Mapping limitations: ecological fallacy (area-level patterns applied to individuals); MAUP (modifiable areal unit problem) - results change with the size and shape of units; edge effects at jurisdiction boundaries; geocoding error; feedback loops - patrolling a hotspot generates more recorded crime there, which reinforces the hotspot.
• Privacy and rights: victim identity must not be mappable to a household (aggregate, jitter the points, or publish at beat level); disclosure of a rape victim's identity is an offence (BNS provision replacing IPC section 228A); POCSO bars identity disclosure. Data protection: the Digital Personal Data Protection Act, 2023 is enacted and the DPDP Rules, 2025 notified, but the Data Fiduciary duties (notice, consent, breach intimation, erasure) and the omission of IT Act section 43A apply only from 13 May 2027; until then IT Act section 43A and the SPDI Rules, 2011 govern sensitive personal data. Law-enforcement processing has statutory exemptions - the exact scope is for the notified provisions.
• Transparency: CII is a public document; the Right to Information Act, 2005 is the route for station-level data not published. NCRB does not publish caste- or religion-wise victim data for general crime heads; caste appears only in SC/ST atrocity heads.
• Evidence angle: a map or a CCTNS extract produced in court is electronic evidence - admissibility now runs through the Bharatiya Sakshya Adhiniyam, 2023 (which replaced the Indian Evidence Act, 1872 from 1 July 2024) and requires the prescribed certificate for electronic records; the BSA section numbers are not the Evidence Act's numbers shifted by a fixed amount, so cite the rule rather than a number unless certain.

Reading and using a CII table - step by step

1. Identify the table universe: all-India, State/UT, or the 19 metropolitan cities (cities of population 2 million and above as per the latest census used by NCRB) - city and State figures must never be added.
2. Check the crime head and the law (IPC/BNS head or which SLL), and whether the table is cases, victims, or persons arrested.
3. Note the year of registration and any footnote on revised or non-comparable figures.
4. Pick the right denominator: total population for general crime; female population for crime against women rates in some tables; child population for crimes against children; number of vehicles for vehicle theft risk.
5. Compute rate = cases × 100,000 / population; compute share = State cases / all-India cases × 100.
6. Separate police-stage indicators (charge-sheeting, pendency of investigation) from court-stage indicators (conviction rate, trial pendency) before drawing conclusions about "police performance".
7. For trend, use rates over at least 5-10 years, flag definitional breaks, and prefer CAGR or a fitted trend over a two-year comparison.

Key points

• NCRB - MHA bureau set up in 1986; publishes Crime in India, ADSI, Prison Statistics India; compiles from SCRBs, does not record crime itself.
• Crime in India - annual report on cognizable crime, published since 1953; the standard citation for Indian crime data.
• Cognizable offence - arrest without warrant, FIR registered under BNSS section 173 (former CrPC 154); CII counts these.
• IPC → BNS - the Bharatiya Nyaya Sanhita, 2023 replaced the IPC from 1 July 2024, so crime heads changed and long series are not strictly comparable.
• SLL - Special and Local Laws (NDPS, Arms, Excise, POCSO, IT Act); counts reflect enforcement drives, not incidence.
• Principal Offence Rule - only the most heinous offence in an incident is counted, which undercounts lesser offences.
• Crime rate - cases per 100,000 population; the only fair basis for comparing States of different size.
• Projected population - rates for recent years use Technical Group projections because the Census due in 2021 was deferred (next Census reference date 1 March 2027; 1 October 2026 in snow-bound areas).
• Charge-sheeting rate vs conviction rate - the first measures police investigation, the second measures the trial outcome; never equate them.
• Dark figure of crime - the gap between crime committed and crime recorded; measured only by victimisation or self-report surveys, which India lacks nationally.
• Lalita Kumari (2014) - Supreme Court made FIR registration mandatory in cognizable cases, which raises recorded crime without raising actual crime.
• CCTNS and ICJS - the digital spine: station-level records, and linkage of police, courts, prisons, prosecution and forensics.
• Hotspot - a statistically significant spatial cluster (Getis-Ord Gi\*, Moran's I), not just a dark patch on a density map.
• Kernel density estimation (KDE) - smoothed surface of crime intensity; the commonest hotspot output in GIS.
• MAUP and ecological fallacy - map results change with unit size, and area patterns must not be read onto individuals.
• Routine activity theory - offender, target, absent guardian converging in space and time: the theory behind mapping.
• Repeat/near-repeat victimisation - raised short-term risk at and near a victimised target; justifies hotspot patrolling.
• Victim privacy - sexual-offence and POCSO victim identities cannot be disclosed or mapped to a household; publish at beat or ward level.

Formulas and examples

Example 1 - Crime rate and why absolute numbers mislead. State A registers 2,40,000 cognizable IPC/BNS cases with a projected population of 12.0 crore. State B registers 1,80,000 cases with a projected population of 3.6 crore.

• State A rate = 240,000 × 100,000 / 120,000,000 = 200 per lakh.
• State B rate = 180,000 × 100,000 / 36,000,000 = 500 per lakh.
• State A has 1.33 times the cases but State B's crime rate is 2.5 times higher. Conclusion: rank on rate, and remember neither tells you about unreported crime.

Example 2 - Charge-sheeting, conviction and pendency from one head. In a district, 1,200 theft cases were registered during the year; 100 were pending from the previous year. Investigation was completed in 900 cases, of which 630 were charge-sheeted and 90 were declared false. Trials were completed in 400 cases, with 160 convictions and 240 acquittals.

• Charge-sheeting rate = 630 × 100 / 900 = 70.0%.
• Conviction rate = 160 × 100 / 400 = 40.0%.
• Cases pending investigation at year end = (1,200 + 100) − 900 = 400; pendency rate = 400 × 100 / 1,300 = 30.8%.
• Note the two rates answer different questions: the police completed and charge-sheeted 70% of completed investigations, while the court convicted in only 40% of completed trials.

Example 3 - Crime against women: choice of denominator. A State records 24,000 cases of crime against women. Total projected population 6.0 crore, of which females are 2.91 crore.

• On total population: 24,000 × 100,000 / 60,000,000 = 40.0 per lakh.
• On female population: 24,000 × 100,000 / 29,100,000 = 82.5 per lakh.
• The second is more than double. Any comparison with another State must use the same base - a classic exam trap.

Example 4 - Trend: two-year jump versus CAGR. Cyber-crime cases in a State: 1,250 in Year 0 and 4,000 in Year 5.

• Five-year percentage change = (4,000 − 1,250) × 100 / 1,250 = 220%.
• CAGR = [(4,000 / 1,250)^1/5 − 1] × 100 = (3.2^0.2 − 1) × 100. ln 3.2 = 1.16315; 1.16315/5 = 0.23263; e^0.23263 = 1.26193 → ≈ 26.2% per year.
• A 220% headline is the same thing as 26% a year compounded; and part of the rise is better registration and new crime heads under the IT Act, not necessarily more offending.

Example 5 - Hotspot sanity check with the nearest neighbour index. A police station area of 20 km² has 50 recorded snatchings; the observed mean nearest-neighbour distance is 140 m.

• Density n/A = 50 / 20,000,000 m² = 2.5 × 10⁻⁶ per m²; √(2.5 × 10⁻⁶) = 1.5811 × 10⁻³.
• Expected distance = 0.5 / 1.5811 × 10⁻³ = 316.2 m.
• NNI = 140 / 316.2 = 0.443 → strongly clustered, so hotspot patrolling is justified. Had NNI been ≈ 1, the dark patch on the KDE map would be an artefact of the bandwidth.

Where it is used

• Police departments and SCRBs: crime analysts and statistical assistants compile CII returns, run CCTNS dashboards, prepare beat-wise hotspot maps for night deployment, and brief at weekly crime review meetings.
• Central agencies and research bodies: NCRB, BPR&D (research studies, *Data on Police Organisations*), NIC/ICJS teams, Lok Nayak-type institutes and university departments; NHRC and NCW use the same data for rights monitoring.
• Policy, budget and judicial use: Finance Commission and State planning for police modernisation, the Nirbhaya Fund safety-audit and dark-spot projects, Safe City schemes in metros, and Supreme Court/High Court directions that cite CII figures on pendency and custodial deaths.
• Road safety and urban planning: accident black-spot mapping on National Highways (MoRTH/NHAI), ward-level safety planning and street-lighting under Smart Cities missions use the same GIS methods.
• Private sector: insurance underwriting and claims fraud, retail shrinkage and store-location risk, banking ATM and cash-logistics route risk, private security deployment, and risk consultancies that build India city-risk indices.
• Media, NGOs and advocacy: data journalism units, PRS and India Justice Report-type compilations, Dalit and women's rights organisations using atrocity and crime-against-women tables with RTI for station-level gaps.

Interview questions

• Q: What exactly does NCRB's Crime in India report cover? A: Cognizable crimes registered by police under the IPC/BNS and Special and Local Laws across all States and UTs, with cases, victims, persons arrested, disposal by police and by courts; it does not cover unreported crime or ordinary non-cognizable offences.
• Q: Why can two consecutive years of NCRB data be non-comparable? A: Definitions and crime heads change - the shift from IPC to BNS from 1 July 2024, new heads such as POCSO or cyber offences, amended definitions after the 2013 criminal law amendment - and registration practice itself changes after court directions or helpline drives.
• Q: What is the Principal Offence Rule and what does it do to the numbers? A: When one incident involves several offences, only the most heinous is counted, so the lesser offences in the same incident never appear and total offence counts are understated.
• Q: Crime rate or absolute numbers - which do you report, and why? A: Crime rate per 100,000 population, because absolute counts simply track population size; but I state the denominator used and note that recent-year denominators are projections since the 2021 Census was deferred.
• Q: A State shows a falling crime rate. Is crime falling? A: Not necessarily - recorded crime can fall because of refusal to register, settlement at the station, loss of public confidence or reclassification, so I would cross-check charge-sheeting and pendency, helpline and hospital data, and any survey evidence.
• Q: Difference between charge-sheeting rate and conviction rate? A: Charge-sheeting rate is charge-sheets as a share of investigations completed by police; conviction rate is convictions as a share of trials completed by courts - a high charge-sheeting rate with a low conviction rate points to weak evidence or poor prosecution.
• Q: What is the dark figure of crime and how would you estimate it in India? A: The crime committed but never recorded; estimated through victimisation surveys, self-report studies and triangulation with hospital, insurance and helpline records - India has no regular national victimisation survey, which is the main data gap.
• Q: Which theory justifies hotspot policing? A: Routine activity theory and crime pattern theory - crime concentrates where motivated offenders, suitable targets and absent guardians converge along the paths and nodes people routinely use - supported empirically by repeat and near-repeat victimisation.
• Q: What is MAUP and why should a crime analyst care? A: The modifiable areal unit problem: results change with the size and boundaries of the units chosen, so the same data can show or hide a hotspot depending on whether you map by beat, ward, grid cell or police station - report the unit and test alternatives.
• Q: How do you know a dark patch on a KDE map is a real hotspot? A: Test it statistically - Getis-Ord Gi\* z-scores above about 1.96, global Moran's I, or a nearest-neighbour index well below 1 - and check that the pattern survives a change of bandwidth and of time window.
• Q: What privacy and legal limits apply when you publish a crime map? A: Points must not identify a victim's household; sexual-offence and POCSO victim identities cannot be disclosed; publish aggregated or jittered data at beat or ward level, and treat sensitive personal data under IT Act section 43A and the SPDI Rules until the DPDP Act's Data Fiduciary duties commence on 13 May 2027.
• Q: What is CCTNS and how does it change crime analysis? A: The Crime and Criminal Tracking Network and Systems digitises FIRs, case diaries and final reports at police stations and links them through the Inter-operable Criminal Justice System to courts, prisons, prosecution and forensics - giving near-real-time, geocodable, station-level data instead of annual aggregates.
• Q: Give one way hotspot policing can bias the data it feeds on. A: Extra patrolling in a hotspot produces more detections and more recorded crime there, which makes the hotspot look hotter and attracts further patrolling - a feedback loop that must be controlled by comparing with victimisation or calls-for-service data.

Common mistakes

• Comparing States or cities on absolute case counts instead of crime rates, or adding metropolitan-city figures to State figures (double counting).
• Treating NCRB data as a measure of crime committed rather than crime recorded, and ignoring the dark figure and burking.
• Confusing charge-sheeting rate with conviction rate, or computing conviction rate on cases registered instead of trials completed.
• Mixing the denominator - using total population for crime-against-women rates in one State and female population in another - and forgetting that recent denominators are projections.
• Quoting IPC section numbers and the CrPC as current law: the BNS, BNSS and Bharatiya Sakshya Adhiniyam replaced the IPC, CrPC and Evidence Act from 1 July 2024, and series across that break are not strictly comparable.
• Calling any dark area on a density map a hotspot without a significance test, and not reporting the bandwidth or the areal unit (MAUP, edge effects, ecological fallacy).
• Plotting exact incident points for sexual offences or POCSO cases, or publishing maps that can identify a victim's house.
• Reading a large percentage jump in SLL crimes (NDPS, Excise, Arms) as rising criminality when it reflects an enforcement drive.

Related topics

Criminology careers: police, CAPF, correctional services and research · Research methods in criminology: sampling, surveys and measurement · Theories of crime causation: social disorganisation and routine activity · Criminal justice system in India: BNS, BNSS and Bharatiya Sakshya Adhiniyam · Victimology and victim compensation schemes · Cyber crime and digital forensics.