A full account of the Reproductive Stress Index — its construction, data sources, weighting rationale, regression pipeline, and known limitations.
The Reproductive Stress Index (RSI) is a composite score from 0–100 that measures the demographic and structural vulnerability of an Indian state. A higher score means the state is better positioned to manage fertility decline — lower means it faces compounding risks.
Population receives the highest weight (35%) because RSI is fundamentally a fertility stress index — the rate and trajectory of TFR decline is the primary signal. Economic capacity (30%) determines whether a state can absorb demographic change through pension systems, healthcare, and migration. Social infrastructure (20%) captures long-run resilience — educated women and strong healthcare systems buffer demographic shocks. Geopolitical stability (15%) is the softest signal and receives the lowest weight deliberately — it modulates the score without dominating it.
Constructed from three sub-indicators:
Constructed from three sub-indicators:
Constructed from NFHS-5 direct survey data:
Equal-weighted composite of four factors (Scheme C):
Scheme C (equal weight) was selected after testing four weighting configurations. Spearman rank correlations between all schemes exceeded 0.986 — scheme choice does not materially affect rankings.
Three models were fitted independently for all 36 states using NFHS-3, NFHS-4, and NFHS-5 as the three observation points. The best-fit model per state was selected by lowest RMSE. Polynomial regression won 32 of 36 states.
Assumes TFR falls at a constant rate each year. Interpretable — slope directly gives annual decline. Fails for states with non-monotonic trajectories (Kerala, Tamil Nadu recovery).
Degree-2 parabola captures curvature — deceleration, acceleration, and recovery. The a coefficient is the key signal: a > 0 indicates a recovery curve; a < 0 indicates accelerating decline. With 3 data points, polynomial achieves perfect in-sample fit (RMSE = 0).
Demographically most realistic — TFR asymptotically approaches a floor f and never crosses it. Parameter k is the decay rate; half-life = ln(2)/k. Fitted via scipy curve_fit (Levenberg-Marquardt). Used for 95% bootstrap confidence intervals on all projections.
95% confidence intervals are computed via bootstrap resampling (300 iterations). At each iteration, residuals from the fitted model are resampled with replacement and added to the predicted values to generate a perturbed observation set. The model is re-fitted on this set, producing a new projection curve. The 2.5th and 97.5th percentiles of the 300 curves form the CI band. Wide CI bands indicate states where the trajectory changed direction between survey rounds — the model is least reliable for these states.
All projections extend to 2045. Confidence in projections degrades significantly beyond 2035 given only three observation points. Milestone years (hitting replacement at 2.1, hitting severe decline at 1.5) should be read as order-of-magnitude estimates, not precise forecasts. The NFHS-6 survey (~2026) will provide a fourth data point that will substantially tighten projection uncertainty.
Population, Economic, and Social dimensions with placeholder geopolitical scores (flat 50.0 for all states). Social dimension used HDI proxies rather than direct NFHS-5 survey data. Global comparison not yet included.
Social score rebuilt from direct NFHS-5 variables: female literacy, institutional delivery rate, contraceptive prevalence, and women's education attainment. Global Phase 2 analysis added — India's TFR trajectory compared to South Korea, Japan, Israel, and global average. Convergence speed analysis added.
Geopolitical score replaced placeholder with four-factor composite: border conflict, insurgency intensity (SATP), disaster vulnerability (NDMA), and communal tension (NCRB). Four weighting schemes tested (A–D); Scheme C (equal weight) selected after Spearman correlation analysis confirmed scheme choice does not materially affect rankings (all inter-scheme r > 0.986). Time series regression pipeline added: Linear, Polynomial, and Exponential Decay models fitted for all 36 states with 95% bootstrap confidence intervals and projections to 2045.
All regression models are fitted on three observations (NFHS-3, 4, 5). Polynomial regression perfectly interpolates three points by construction — RMSE of zero does not indicate model quality, only that a degree-2 polynomial has exactly three parameters for three points. Projections beyond 2030 carry substantial uncertainty.
Maharashtra's RSI score masks the gap between Mumbai (TFR ~1.4) and Vidarbha (TFR ~2.3). District-level analysis would reveal substantially more variance. The current dataset treats each state as homogeneous, which is a significant simplification for large, diverse states.
India's 2021 Census was delayed. All population-based calculations (age structure, dependency ratios, per capita metrics) use 2011 Census data with linear projections. States with high out-migration or urbanization (UP, Bihar) may have significantly different actual population structures than the projections suggest.
The border conflict, insurgency, disaster, and communal sub-scores are scored on a 0–100 scale using referenced data sources but involve judgment in aggregation. The equal-weighting (Scheme C) was chosen for defensibility, not because equal weights are demonstrably correct. A different researcher might score J&K's border exposure differently.
RSI measures correlation between structural factors and demographic vulnerability. It does not claim that high GDP causes higher RSI, or that insurgency causes low fertility. The index is descriptive and comparative — it ranks states on observable conditions, not on underlying mechanisms.
Kerala and Tamil Nadu show TFR rising from NFHS-4 to NFHS-5. The polynomial model projects this recovery continuing. However, one data point of increase is insufficient to confirm a structural recovery — it may reflect survey noise, boundary changes, or a temporary reversal. NFHS-6 data will be critical for validation.