Evidence in Indian policymaking: data systems, evaluation and honesty about failure

"Evidence-based policy" has become one of those phrases every Indian ministry now uses in its annual reports, alongside "outcome-oriented" and "data-driven," often with limited attention to what the phrase actually requires institutionally. It requires, at minimum, three things: statistical systems robust enough to be trusted, an institutional culture willing to commission and publish evaluations even when they find a favoured scheme has failed, and enough separation between the agencies producing data and the political leadership whose performance that data measures. India's record on all three has been genuinely mixed, and the gaps are worth stating precisely rather than gesturing at vaguely.

A real capacity, unevenly protected

It would be wrong to begin this discussion without acknowledging what does work. India's statistical infrastructure — anchored by the National Sample Survey Office, now part of the National Statistical Office, and the decennial Census — has produced, over seven decades, some of the more extensive household-level datasets available anywhere in the developing world, covering consumption expenditure, employment, health and a wide range of other indicators. The methodological rigour of the NSS's sampling design has been internationally respected, and Indian statisticians trained within this system have contributed meaningfully to global survey methodology debates.

That capacity has, however, been repeatedly tested by political pressure at the point where data becomes inconvenient. The 2019 episode in which the government withheld the 2017-18 Periodic Labour Force Survey results, which reportedly showed unemployment at a multi-decade high, before eventually releasing the data after it was leaked and became a matter of public controversy, remains the starkest recent example. Two independent members of the National Statistical Commission resigned at the time citing exactly this concern, over both the labour survey delay and unresolved questions about back-series GDP data methodology. Whatever the eventual technical resolution of those specific disputes, the sequence damaged public trust in the release process itself, since a delay coinciding precisely with an unfavourable finding invites the reasonable inference that release timing was being managed for political convenience rather than statistical readiness.

The evaluation gap

A separate, less publicly visible problem concerns programme evaluation: the systematic assessment of whether a government scheme, once implemented, achieved what it promised. India runs an enormous number of centrally sponsored schemes, and the NITI Aayog's Development Monitoring and Evaluation Office, along with various ministry-level evaluation units, does commission studies of scheme performance. But a persistent pattern, documented by researchers who have tried to access these evaluations through Right to Information requests, is that unfavourable evaluations are far less likely to be published or publicised than favourable ones, and in some cases are not released at all, remaining internal documents cited obliquely, if at all, in subsequent budget or policy announcements.

This is not unique to India — bureaucracies everywhere have an incentive to suppress evidence of their own programmes' shortcomings — but the absence of a strong countervailing institutional force, such as a statutorily independent evaluation office with guaranteed publication rights comparable to the Government Accountability Office's role in the United States federal system, means the correction mechanism in India relies heavily on investigative journalism, opposition parliamentary questions, and the occasional Comptroller and Auditor General report, all of which are slower and less systematic than a dedicated evaluation function would be.

When evidence is used well

The exceptions are instructive. The Ayushman Bharat health insurance scheme's implementation has been subject to fairly regular, published claims-data analysis, including third-party studies examining fraud patterns and hospital empanelment issues, which has fed back into programme design changes over successive years. The direct benefit transfer architecture built on the JAM trinity of Jan Dhan accounts, Aadhaar and mobile numbers has similarly generated a reasonably rich evidence base, partly because DBT's leakage-reduction claims became politically valuable to demonstrate, giving the government an incentive to publish supportive data rather than suppress it — an important reminder that the willingness to generate and release evidence often tracks whether the expected finding is politically convenient, in India as everywhere else.

The counter-argument: perfect data can also become an excuse

Critics of the evidence-based policy framework, including some experienced Indian administrators, make a fair point in response: waiting for methodologically perfect data before acting can itself become a bureaucratic excuse for inaction or for delaying difficult decisions indefinitely, particularly in a country where administrative capacity to conduct rigorous randomised evaluations at scale, of the kind associated with the Abdul Latif Jameel Poverty Action Lab and its India-based studies, remains concentrated in a small number of institutions and geographies. Some decisions, especially in fast-moving crises such as pandemic response, genuinely have to be made on incomplete evidence, and a policy culture excessively deferential to formal evaluation risks paralysis dressed up as rigour. This is a legitimate caution, but it does not excuse the withholding of evidence that already exists and has already been collected at public expense, which is a different failure from the absence of evidence in the first place.

What would strengthen the system

A more resilient framework would separate data production from political oversight through statutory guarantees — something closer to the operational independence the Reserve Bank of India enjoys on monetary data, extended to the National Statistical Office's release calendar, which should be fixed, published well in advance and legally difficult to alter for reasons other than a documented, published methodological concern. It would also establish a genuinely independent programme evaluation office, with guaranteed timelines for public release of commissioned evaluations regardless of their findings, insulated from the ministries whose schemes are being evaluated. And it would build a stronger norm, cultivated as much by media and civil society as by government, that a published evaluation showing a scheme underperforming is a sign of institutional health rather than political embarrassment. India has the technical capacity to be a global leader in public-sector data and evaluation. What it has lacked, at crucial moments, is the institutional courage to let that capacity operate without political interference in the moments it matters most.