Traps in Indian data
Anyone can find a number. The harder part is knowing what it does not say. These are the mistakes that catch experienced people, in newspapers, in submissions to government and in published research. None needs bad faith. Each only needs someone to skip the notes.
The traps
The 2026 rebasing broke every long series
In 2026 India moved GDP and industrial production to a 2022-23 base and consumer prices to 2024=100. A matching back series was expected by December 2026. Until it arrives, any chart running from 2011-12 to today joins two series that do not match. The official note on first-quarter 2026-27 growth says as much: it can only be compared with the new-series figure for a year earlier. Rebasing also moves every ratio set against GDP, such as debt, deficit or health spending, without a rupee changing hands.
What to do: Say which series a number comes from. For a long run, stay on the old series and stop at 2022-23, or wait for the back series.
In Sarani: New GDP series, New CPI series, New IIP series
A change in the law can look like a change in crime
Crime in India 2024 showed registered cognisable crime falling by about 6 per cent. It was also the first full year under the Bharatiya Nyaya Sanhita, which made simple hurt non-cognisable. Cases of hurt fell sharply as a result. The same report shows cybercrime rising past one lakh cases for the first time.
What to do: Before reading a year-on-year change as a change in behaviour, check whether the offence was redefined in that year.
In Sarani: NCRB
Registered crime is not crime
NCRB counts cognisable offences recorded in first information reports. Non-cognisable offences are left out. Under the principal offence rule, only the most serious offence in an FIR is counted, so a rape that ends in murder appears once, as murder. NCRB's own disclaimer says counts under each head may be understated. A state that records more crime may simply record more honestly.
What to do: Write registered cases, not crime. Where your argument depends on how much crime actually happens, look for a victim survey as well.
In Sarani: NCRB
One labour survey, two very different answers
PLFS measures work over two periods. Usual Status looks back a year and counts anyone with a real spell of work. Current Weekly Status looks back a week. The yearly measure always gives higher participation, above all for women, because it picks up seasonal and occasional work. Both figures are correct. Setting one against the other is not.
What to do: Put the measure, the period and the age group in the same sentence as the number.
In Sarani: PLFS
Forest cover is not forest
The Forest Survey counts any patch above one hectare with more than ten per cent tree canopy, whoever owns it and whatever grows there. Plantations, orchards and coffee shade trees all count. Cover can rise while natural forest shrinks. The baseline also shifts as satellite images get sharper.
What to do: Look at very dense, moderately dense and open forest separately, and read the change tables as well as the headline.
In Sarani: FSI / ISFR
Planned money and spent money are different numbers
A budget shows next year's Budget Estimate, this year's Revised Estimate and an earlier year's Actuals side by side. Comparing this year's estimate with last year's revised figure often shows an increase that disappears once Actuals arrive, about two years later.
What to do: Compare like with like. To say money was spent, use Actuals.
In Sarani: indiabudget.gov.in, CAG
Registrations are not people served
Most scheme dashboards show everyone ever registered. A registration is not an active user, an eligible person or someone who received a benefit. These totals never fall, so they look like steady progress whatever is happening on the ground.
What to do: Look for a number on payments made or services used. Where only registrations exist, call them registrations.
Surveys and administrative counts will disagree
NFHS measures stunting in a random sample of children, weighed by trained teams. Poshan Tracker records children who came to an anganwadi and were measured there. The groups, conditions and incentives differ, so the numbers differ. The same gap shows up between NFHS and HMIS on births in hospitals.
What to do: Use surveys for how common something is. Use administrative data for coverage and flow within a programme. Do not stitch the two into one series.
In Sarani: NFHS, Poshan Tracker, HMIS
A 2011 base is now fifteen years old
Until Census 2027 tables are out, every per-person rate in India rests on projections from 2011. Hundreds of districts have been created since then, and cities have grown well past their 2011 limits. Crime rates, doctors per person and spending per person all carry that error.
What to do: For district work, check whether the district existed in 2011. For city work, treat the 2011 population as a floor.
In Sarani: censusindia.gov.in, Census 2027
District names are not identifiers
India has more than one Aurangabad, Hamirpur and Bilaspur. Spellings change between ministries, languages and years. Joining two district tables on names will drop some rows and mismatch others without warning. The Local Government Directory gives stable codes down to the village and records every split and merger.
What to do: Convert everything to directory codes first. When a district was split, decide whether to add the new parts back to the parent or drop those years, and say which you did.
In Sarani: LGD
Two consumption series, two poverty stories
The 2023-24 consumption survey reports spending per person both with and without the value of free items received through welfare schemes. The gap between the two is wider in poorer states. Neither is wrong, but a comparison with 2011-12, which did not count free items this way, is only a rough one.
What to do: Name the series you used, and describe any comparison with 2011-12 as indicative.
In Sarani: HCES
Air quality numbers depend on which monitors you count
The pollution board's network mixes stations that report every hour with manual stations that sample a few days a month. Some cities are covered densely and most districts not at all. A yearly average from a station that went quiet through the monsoon is not comparable with one that ran all year.
What to do: Check how many valid days sit behind any yearly figure before you compare places.
In Sarani: CPCB
Groundwater figures are modelled, not measured
The groundwater board estimates how much is drawn from recharge models and indirect signs such as electricity connections and crops, not from meters. Labels such as over-exploited are thresholds applied to those estimates, and areas are reassessed between editions.
What to do: Use it to see patterns and direction. Do not treat an area moving from one label to another as a measured event.
In Sarani: CGWB
Pending cases do not measure delay
The judicial data grid counts cases pending, filed and decided. Pending cases are a stock that moves with new filings, vacancies among judges, how case types are defined and how each court enters data. Pending cases can rise while cases are being decided faster.
What to do: Pair pending cases with the age of cases and with how many judge posts are actually filled.
In Sarani: NJDG
Potential, capacity and generation are three different things
Energy statistics report assessed potential, installed capacity and power actually generated. A new assessment method can raise potential overnight. Installed renewable capacity also produces far less power per megawatt than coal, so capacity alone overstates the shift.
What to do: Name which of the three you mean. For an argument about cutting emissions, generation is usually the number that counts.
In Sarani: Energy Statistics India
Voter numbers are moving for a reason
The Special Intensive Revision asks every registered voter to confirm their details, state by state, and removes entries for people who have died, moved or been listed twice. A fall in registered voters over 2025 and 2026 largely reflects this revision, not a change in who is eligible to vote.
What to do: Compare electoral rolls only from before or only from after the revision in a given state, and say which you used.
In Sarani: ECI
If you remember one thing
Put the caveat in the sentence, not the footnote. Registered cases of crime against women fell is a sentence you can defend. Crime against women fell is not, and the difference is one word.