How the evidence is cross-checked
and how the three indices are built
This study is descriptive. It sets the record of the UPA decade (2004-2013) beside that of the NDA period (2014-2026) on published statistics, and adds three indices built from documented events and published third-party series. It identifies no causal effects. The method is designed for an economy in which the reliability of official data is itself a research question.
The SSI is a weighted sum of graded suppression severities (0-10 scale). A persistence (scar) term makes a resolved suppression decay gradually towards zero. It is computed identically for both eras in data/compute_indices.py. The UPA decade records no qualifying event, so its SSI is 0.
| Component | Weight | Trigger / event |
|---|---|---|
| Census delay | 2.5 | 2021 census postponed in 2020 with no date (severity 1.0 for 2020-2025); houselisting April-September 2026 and enumeration notified for February 2027, so 2026 is scored 0.3 as a delay |
| Consumption survey | 1.5 | 2017-18 survey scrapped in November 2019 (scored from 2019), leaving an 11-year gap to HCES 2022-23 |
| Employment data | 1.0 | PLFS 2017-18 report approved in December 2018 and withheld until 31 May 2019 (scored from 2018) |
| GDP back-series | 1.5 | Disputes over the 2011-12 base (0.7 for 2015-2017), then the back-series shelved in 2018 (1.0 from 2018) |
| Institutional independence | 1.5 | NSC resignations (2019) + permanent NSSO→NSO merger |
| Mortality undercount | 2.0 | COVID deaths: official ~0.5M against 3-5M excess-mortality estimates |
The FCI measures the shift of fiscal authority from the states to the Union. It averages the components published for each year. Four are published fiscal ratios, each min-max normalised over the years in which it is published; two are coded:
The FCI averages 0.30 under UPA and 0.43 under NDA, peaks at 0.87 in 2020 and is 0.35 in 2026. On the four measured ratios alone the UPA years average 0.49 and the NDA years 0.45, so the higher NDA score comes from the GST and the 2020 borrowing conditions. The index was rebuilt from these published figures on 10 October 2026.
The DQI uses a geometric mean of four published third-party measures to penalize weakness in any dimension:
The comparison is between period averages. For each measure the study takes the mean of the UPA years and the mean of the NDA years that have a published figure, and reports which is better. Nine measures are compared on a consistent basis across both periods: GDP growth, CPI inflation, the fiscal deficit, the top 1% and bottom 50% income shares, the press-freedom rank and the three indices. The NDA period is worse on eight of the nine; inflation is the exception.
Series that begin with the Periodic Labour Force Survey in 2017-18 (unemployment on the PLFS basis, youth and graduate unemployment, women's labour force participation, formal employment) have no UPA-period counterpart on the same basis. They are reported for the NDA years and left out of the cross-era comparison.
Employment elasticity is not estimated here. Appendix A5 of the technical appendix reports the published estimates with their sources and periods.
The study runs no difference-in-differences, synthetic-control or structural-break analysis. An earlier version of this page said it did; that was wrong, and the description has been removed.
Each series uses the publisher of that statistic: MoSPI for the labour force surveys, prices and national accounts; the Union Budget documents, the Finance Commission reports and the RBI for fiscal data; the World Inequality Lab for income shares; V-Dem, Freedom House and RSF for the democracy measures. Where two official sources differ, the data file says which one is used and why.
GDP measures work inside a production boundary set by the System of National Accounts. The boundary is a convention. Where a household owns the dwelling it occupies, the accounts estimate the rent it would otherwise have paid and add that to output, because a real service is being consumed. The same reasoning is not applied to cooking a meal in that dwelling, minding a child or nursing a parent. Those services enter the accounts at zero.
India already collects the hours. The NSO ran Time Use Surveys through 2019 and 2024. Among those aged 6 and above who do it at all, women spent 289 minutes a day on unpaid domestic services in 2024 and men 88. Among women aged 15-59 who do it, the figure is 305 minutes. For a woman that is roughly 1,758 hours a year, in no employment statistic and no output series. Published valuations of that work span 7.5% to 36% of GDP. The spread comes from the choice of valuation method, and there is little uncertainty about the hours themselves.
No figure in this assessment is adjusted for it. Every growth rate is the official one on the official boundary. Full treatment: someperspective.info/care and Section 6.11 of the research paper.
Three checks are reported in the technical appendix (section A7). The SSI is recomputed with every weight moved by 20% either way. The FCI is recomputed without the coded borrowing component, which moves the 2024 value from 0.44 to 0.53. The DQI is recomputed as an arithmetic mean, which moves the 2024 value from 0.28 to 0.33. None of the three changes the direction of the cross-era comparison.
The study reports no bootstrap intervals, Monte Carlo simulations or external-validation correlations. An earlier version of this page listed them, with correlation figures that nothing in the repository computed, and they have been removed.
Repository: github.com/Varnasr/someperspective
data.json: every series shown on the site, with a source note for each.data/compute_indices.py: the SSI, FCI and DQI, their inputs and the sources of those inputs.data/SP_masterdataset.csv: the year-by-year dataset, written from the two files above by tools/build_master.py.replication_code.py and replication_code.R: recompute the cross-era comparison from the dataset.paper/paper.md: the source of the working paper.git clone https://github.com/Varnasr/someperspectivepython3 data/compute_indices.pypython3 tools/build_master.pypython3 replication_code.py (standard library only; matplotlib is optional) or Rscript replication_code.R (base R)
Technical methodology | Some Perspective
Full research and data: someperspective.info
Last updated: 10 October 2026