The column that inverts the answer
India's unemployment rate in 2025, broken down by social group, runs like this: Scheduled Tribes 2.1%, Other Backward Classes 3.0%, Scheduled Castes 3.3%, and the residual “Others” category 3.7%.
Taken by itself that table says Scheduled Tribes are doing better in the labour market than anyone else, and that people outside the three reserved classifications are doing worst. Almost nobody working in India would recognise that as a description of the country, and it would make a strange basis for a programme.
The participation column runs the other way. 73.1% of Scheduled Tribe adults are in the labour force, against 53.9% of “Others” — a gap of twenty points, several times anything in the unemployment column.
PLFS calendar-year 2025, age 15 and over, usual status.
What the rate is actually counting
Being counted as unemployed requires two things at once: not having work, and looking for it. Looking for work requires a household that can manage without your earnings while you look. Where that is not possible, people take whatever is available rather than search, and the recorded rate stays low.
So unemployment barely varies. Across every social group in a country of 1.4 billion people it moves between 2.1% and 3.7%. Participation, across the same four groups, moves twenty points.
Among women the pattern is starker. Scheduled Tribe women have the highest participation in the data, 62.6%, and the lowest unemployment, 1.7%. Women in the “Others” category have the lowest participation, 31.0%, and the highest unemployment, 4.7%. Those are the same women in both figures.
The two rates are not shares of the same population, which is the part a dashboard rarely shows.
Two rates, two different denominators.
A rule for building explorers
We run thirteen data explorers on this site, and some version of this problem turned up in all of them, so we settled on a rule: don't show a headline number without the number that changes its meaning.
On the PLFS state rankings that means 95% confidence intervals, drawn as whiskers. Where two states' whiskers overlap heavily the difference between them may not be real, and a ranking that hides this encourages a sentence the data will not support.
On the Union Budget explorer it means putting 2024-25 actual spending next to the 2025-26 budget, its mid-year revision and the 2026-27 budget. Interest on past borrowing is the largest single head, at ₹14.04 lakh crore, more than defence, education and health together. The page says outright that this is debt service rather than a programme, because the bar chart on its own would suggest otherwise.
On the Energy explorer it means showing assessed renewable potential, installed renewables, and the ratio between them. Rajasthan has more installed renewable capacity than any other state, 26,693 MW, ahead of Gujarat, and has built about 6% of what it could. Punjab has roughly a thirteenth of Rajasthan's installed capacity and has built about 24% of its own potential. Ranking by capacity and ranking by that ratio give very different orders.
And it means keeping the previous round alongside the current one: NFHS-5 against NFHS-4, ASER across a decade, the Happy Planet Index across twenty years. India ranked 11th in the world on that index in 2006 and 65th in 2025, over a period when life expectancy rose by seven years. One ingredient moved, and the only way to see which is to look at the ingredients.
The half that means shipping gaps
The harder part of the rule is saying what a page cannot tell you.
Budget heads get renamed, split and merged between years. Where that happened, the budget explorer shows a blank cell and no change figure, rather than a percentage computed across a definition change. A blank there means the budget changed what it published, not that spending stopped.
The IHDS explorer shows only the states its source names, because the full 36-state table sits inside a paywalled paper. The states in between are left out rather than estimated.
The tax-share calculator states what its number is not: taxes on income are around 27% of Union receipts, no particular rupee is marked as yours, and what it gives you is a proportion rather than a trace.
PLFS has no district view because there is no district estimate. The survey is designed to produce state and national figures. A district number could be produced by apportioning, but it would be invented.
The new social-group panel carries no confidence intervals, because MoSPI does not publish relative standard errors broken out by social group. The rest of the page has whiskers and this part does not, with a line explaining why. Drawing them anyway would have looked tidier and meant less.
What it cost
PLFS moved from a July–June cycle to a January–December one, with a redesigned and much larger sample, so the two series do not form a continuous run. The straightforward option was to substitute the new figures and let the trend line continue. Instead the social-group panel sits on the calendar-year series and is labelled as a different series from the one the rest of the page uses. The page is worse-looking for it.
We also got a number wrong on our own site. The count of Data Notes appeared as six in one place and seven in another when there were eleven, because no canonical count existed for that content type and nothing checked it. It is filed as issue #1020 and fixed, and the key is now in the count guard that runs on every push. Anything nothing checks will drift, our own figures included.
The habit
All thirteen explorers are free and source-linked, at the Data Room. If there is one thing worth taking from this, it is to find the second column before putting a rate into a proposal, a board paper or a press release. It is usually a single column, usually printed in the same table, and leaving it out is how a programme ends up aimed at the wrong households.