The Custody of 6.1 Per Cent

On Monday 28 January 2019, the two non-official members of the National Statistical Commission resigned. P.C. Mohanan, a career statistician of more than thirty years standing who was acting as chairman, and J.V. Meenakshi, professor at the Delhi School of Economics, had both joined in June 2017 and had terms running to June 2020. Their departure left the Commission with two people on it: Pravin Srivastava, the Chief Statistician, and Amitabh Kant, chief executive of NITI Aayog. Three of the seven seats were already vacant.¹

Three days later, Business Standard published the thing they had resigned over. The first annual report of the Periodic Labour Force Survey, covering July 2017 to June 2018, put the unemployment rate at 6.1 per cent, the highest since 1972-73, when the employment and unemployment survey series began. The comparable figure from the last quinquennial round, in 2011-12, had been 2.2 per cent. The labour force participation rate had fallen from 39.5 per cent to 36.9 per cent over the same period. The Commission had approved the report in December 2018. It had not been released.²

That is the whole of the episode in its bare form, and it is worth stating what did not happen. No number was altered. No statistician was instructed to recalculate. The report that finally appeared on 31 May 2019 carried the same estimate as the report that had been withheld in December. What was exercised over those five months was custody rather than authorship, and custody turned out to be enough.

Four positions at once

The government's response to the leak is instructive because it was not one response. It was four, held simultaneously by four ministers and officials, and they were mutually inconsistent.

NITI Aayog and Arun Jaitley, the finance minister, described the report as a draft. Rajiv Kumar, vice-chairman of NITI Aayog, said the four-decade high was unverified. Amitabh Kant, who sat on the Commission itself, said comparing the older employment surveys with the new one was erroneous and that the report did not have his approval. Dharmendra Pradhan, then petroleum minister, said the report did not exist.³

Mohanan's answer to the first of these went to the question of who owns an official statistic. Once the Commission approves an NSSO report, he told NDTV in February 2019, it is the final report; the government may accept it or reject it, but the government does not approve the statistics.⁴ Whether that is strictly correct as a matter of administrative law is arguable, since the Commission was set up by executive resolution in 2005 rather than by statute and its recommendations bind nobody. But the convention it describes was real, and it had held for decades: the Commission clears a report, and the ministry publishes it within days.

Kant's objection is the more interesting one, because it is not obviously wrong. The Periodic Labour Force Survey did change the sampling design. Where the old quinquennial rounds stratified households at the second stage by per capita consumption expenditure, the new survey stratified by the number of household members who had completed matriculation. If educated households are more likely to report open unemployment, and they are, then a design that oversamples them could raise the measured rate without anything having changed on the ground. That is a serious methodological question, and it was put by a person with the standing to ask it.

There is an irony in the design objection that deserves recording. The Periodic Labour Force Survey exists because a task force headed by Arvind Panagariya, then vice-chairman of NITI Aayog, recommended replacing the five-yearly employment rounds with an annual survey, on the reasoning that policy needed labour market data more often than once in half a decade. The survey that produced the inconvenient number in 2019 was built to a specification NITI Aayog had asked for. Its stratification by matriculation was part of that redesign, adopted in advance and known to everyone who approved it.

It was also put by a person who had not raised it in December, when the Commission approved the report, and who raised it in public only after the number leaked. A methodological objection made at that moment does two jobs at once, and only one of them is statistical.

The question was later settled in the ordinary way. Working through both datasets stratum by stratum, an analysis in the Economic and Political Weekly in 2022 found the estimates internally consistent within each survey's own design and comparable across them.⁵ The jump from 2.2 to 6.1 per cent survived the scrutiny that the objection had invited. By then it was three years old and no longer a political fact.

What the delay bought

The report was published on 31 May 2019. The seventeenth Lok Sabha had been elected, results had been declared on 23 May, and the new council of ministers had been sworn in on 30 May. The survey appeared the following day, unchanged.

The sequence is the finding. If the intention had been to suppress the number permanently, the report would not have been released at all, or would have been released with revisions. If the intention had been to correct a methodological error, the correction would have appeared in the released version. Neither happened. What the delay produced was a general election fought without an official unemployment estimate in circulation, and a published estimate that arrived when it could no longer be argued about.

This is the mechanism worth naming, because it recurs and because it is almost invisible in the published record. A number that is delayed is not a number that has been falsified. Every figure in the eventual report is correct. An archive assembled twenty years from now will show a survey conducted in 2017-18 and published in 2019, with a gap of a few months that reads as administrative delay, and no trace in the tables of what the gap was for.

It is worth being concrete about what the machine does, since the argument was conducted almost entirely in the abstract. For the 2023-24 round, enumerators covered 12,743 first-stage sampling units, 1,01,920 households and 4,18,159 individuals, of whom 3,19,773 were aged fifteen or above.⁶ Urban households are visited four times over the year. Each of the figures argued about in 2019 rests on that fieldwork, conducted by people who had no idea what the aggregate would come to.

The number nobody was arguing about

There is a further point about 6.1 per cent that went almost unremarked at the time, and it turns on the definition being used.

The survey produces two unemployment rates for the same year, on two different reference periods. Usual status counts a person as employed if they worked for a majority of the preceding 365 days, or for at least thirty days in a subsidiary capacity within that year. Current weekly status counts a person as employed if they worked for even one hour in the preceding seven days. The first is the more generous test of employment and therefore yields the lower unemployment rate.

For 2017-18 the usual status rate for persons aged fifteen and above was 6.0 per cent. The current weekly status rate for the same people in the same survey was 8.7 per cent.⁶ The headline figure of 6.1 per cent, the one two Commission members resigned over and four ministers disputed, is the all-ages usual status rate, which is to say the smallest of the plausible numbers the survey produced. Nobody suppressed 8.7 per cent. It sat in the same report, in the next set of tables, and attracted no comment at all.

Both series have fallen since. Usual status unemployment reached 3.2 per cent in 2022-23 and stayed there in 2023-24; current weekly status stood at 4.9 per cent. The relative decline is real on either measure. Which of the two numbers a minister quotes in a press conference is a choice made after the arithmetic is done.

What counts as work

The same definitional gap opens much wider on the other side of the labour market, and this is where the survey's own construction does the most work.

Female labour force participation for ages fifteen and above rose from 23.3 per cent in 2017-18 to 41.7 per cent in 2023-24 on usual status, an increase of more than eighteen percentage points in six years. On current weekly status the same rate rose from 21.1 to 35.6 per cent. The gap between the two measures widened from 2.2 percentage points to 6.1.⁶

The arithmetic of that divergence is not mysterious. A woman enters the labour force under usual status if she performed some economic activity for thirty days out of three hundred and sixty-five. She enters under current weekly status only if she was working in the week the enumerator arrived. A rise concentrated in the usual status series therefore describes women whose recorded economic activity is subsidiary, intermittent, and in a large share of cases unpaid work in a household enterprise. The share of workers classified as self-employed, which includes unpaid family labour, rose from 52 per cent in 2017-18 to 58.4 per cent in 2023-24.

None of this makes the increase false. Those days of work happened, and a survey that failed to record them would be worse than one that does. What it means is that the eighteen-point rise and the eighteen-point rise reported in the newspapers are not quite the same object, and that the distance between them is a property of the definition rather than of the labour market.

What was lost

The Commission that lost two members in January 2019 was the body whose approval had made the report final. Mohanan's account of why he left names the sequence plainly: recommendations not acted on, the Commission bypassed on the GDP back series in November 2018, and then a report cleared and not published.³ He described the last as the first known instance of an NSSO report being held because its findings did not match the government's account of itself.

That is a strong claim and it should be treated as a claim rather than an established fact, since the counterexample would by its nature be unpublished. What can be established is narrower. A survey was completed, approved by the body empowered to approve it, withheld for five months across a general election, and then released without a single figure changed.

The apparatus worked. The report exists, the estimate was correct, and the methodological challenge to it failed on examination. What did not work was the timing, and timing is the part of a statistical system that no committee is charged with defending.

1. The resignations of 28 January 2019 were reported by PTI on 29 January and confirmed by both members; see also the Ministry of Statistics and Programme Implementation clarification of 30 January 2019, which stated that the concerns had not been raised in recent meetings of the Commission. ↩
2. Somesh Jha, 'Unemployment rate at four-decade high of 6.1% in 2017-18', Business Standard, 31 January 2019, reporting from the unreleased annual report of the Periodic Labour Force Survey 2017-18. ↩
3. P.C. Mohanan, interviewed by IndiaSpend, April 2019, on the positions taken by NITI Aayog, the finance minister and the petroleum minister, and on the Commission being bypassed. ↩
4. P.C. Mohanan, interview with NDTV, February 2019. ↩
5. 'Consistency in NSSO Employment-Unemployment Estimates Using NSS 68th Round and PLFS Data', Economic and Political Weekly 57, no. 46 (2022). ↩
6. Ministry of Statistics and Programme Implementation, Periodic Labour Force Survey Annual Report, July 2023 to June 2024, Tables 1, 3, 4 and 6, released 23 September 2024. All figures are for persons aged fifteen and above, all-India, rural and urban combined. ↩