Estimating India's Poor-but-Not-Destitute Population
Estimating India’s “Poor-but-Not-Destitute” Population from HCES 2023-24
Section titled “Estimating India’s “Poor-but-Not-Destitute” Population from HCES 2023-24”A consumption-to-income projection methodology, triangulated across CMIE, PRICE, and WIL savings schedules.
About this paper. This is an internal analysis by Avanti Fellows, shared for transparency. It has not been peer-reviewed. It was prepared with substantial help from AI tools (Anthropic’s Claude), which wrote much of the code and drafted much of the text under the direction of Avanti staff. The numbers are produced by code from the underlying data, and the sources and methods are documented so they can be checked, but errors are possible. Read the findings as preliminary, and contact Avanti Fellows before citing them.
Executive Summary
Section titled “Executive Summary”This paper documents an end-to-end method for estimating how many Indians live in households earning ≤ ₹4 lakh per annum (LPA), starting from the raw NSS Household Consumption Expenditure Survey (HCES) 2023-24 microdata. Because HCES measures consumption but not income, we project income at the household level via the savings identity Income = Consumption / (1 − s), applying three published savings-rate schedules — CMIE-CPHS, PRICE-ICE/RBI-NAS, and the World Inequality Lab (WIL) Bharti-Chancel-Piketty-Somanchi 2024 series — and reporting the simple mid-point across the three as our central estimate.
Headline result (mid-point):
494 million Indians — 41% of the population — live in households earning ≤ ₹4 LPA. Of these, 468 million (38% of the population) are “poor-but-not-destitute” (HH income ₹1.25–4 LPA). At the ≤ ₹8 LPA EWS income cutoff: 967 million Indians, or 79% of the population. Triangulation range across the three schedules: 39%–43% (₹4L) and 78%–82% (₹8L). Aggregate household savings rate at the mid-point: 24% (consistent with RBI/NAS ~22%). Mid-point income Gini: 0.445.
The estimate is geographically concentrated: UP + Bihar alone hold 38% of the segment (≈190M people). Six states — UP, Bihar, MP, Jharkhand, Odisha, West Bengal — account for ~65%.
1. The Question and Why It’s Hard
Section titled “1. The Question and Why It’s Hard”India publishes household consumption data once every 5–10 years (HCES) and labour earnings data annually (PLFS), but neither directly measures total household income — neither the formal-sector salaries nor the informal-sector cash flows nor non-labour income at the household level. Income data exists only in:
- CMIE Consumer Pyramids Household Survey (CPHS) — a panel of ~174,000 households, paid product, with documented under-coverage at both tails (Pais 2021).
- PRICE-ICE 360° — a cross-section of ~200,000 households, last fielded in 2021, with deliberate oversampling of affluent.
- World Inequality Lab (WIL) — a synthetic series combining NSS surveys, PLFS, tax records, and Forbes lists, calibrated to National Accounts aggregates.
Each disagrees with the others on levels but converges on shape in the middle of the distribution. The pragmatic path is to use the most recent and most representative consumption microdata (HCES 2023-24) and project income using all three schedules, then take a mid-point.
2. Data
Section titled “2. Data”NSS Household Consumption Expenditure Survey 2023-24 — 15 level files, ~262,000 sampled households, weighted to ~1.22 billion individuals. Public microdata released December 2024 by MoSPI.
The 15 level files cover:
| Level | Section | Content |
|---|---|---|
| 1 | 1, 1_1 | Household identification + sampling weight |
| 2 | 3 | Person-level demographics |
| 3 | 2 | Household characteristics (size, type, religion, social group, dwelling, fuels) |
| 4, 7, 11 | 4_1, 4_2, 4_3 | Subsidies and online purchases |
| 5 | 5, 6 | Detailed food consumption (item-level, 12.7M rows) |
| 6 | 7 | Clothing & footwear consumption |
| 8 | 8_1 | Other consumables |
| 9 | 9, 10, 11 | Education, medical, transport services |
| 10 | 12 | Rent, taxes, services |
| 12, 13 | 13, 14 | Durables (first purchases, second-hand) |
| 14, 15 | A1-C1, A2-C2 | Section summaries + per-visit monthly consumption totals |
For this analysis we use Levels 1, 3, and 15. Levels 5/8/9/13/14 (item-level) are not needed for the household master.
3. Methodology
Section titled “3. Methodology”3.1 Household master
Section titled “3.1 Household master”We build a one-row-per-household master by joining L1, L3, and L15 on a composite household ID:
HHID = FSU_Serial_No _ Sample_SU_No _ Sample_Sub_Division_No _ Second_Stage_Stratum_No _ Sample_Household_NoFrom L15 we sum MONTHLY_CONSUMPTION_EXP across the three sections that capture distinct expenditure baskets:
- Section A2 — Food (1-week reference period, monthly equivalent)
- Section B2 — Consumables (30-day reference period)
- Section C2 — Durables (365-day reference period, monthly equivalent)
The household total monthly expenditure is A2 + B2 + C2. MPCE is total / hh_size. Survey weight is Multiplier / 100 (NSS convention).
3.2 The consumption pyramid (empirical baseline)
Section titled “3.2 The consumption pyramid (empirical baseline)”Households are binned by MPCE into fixed bands; population shares are weighted by hh_size × survey_weight. The result is robust and assumption-free — it is what HCES directly measures.
| MPCE band (₹/month) | Share of population | Share of consumption |
|---|---|---|
| < ₹1.5k | 0.2% | 0.0% |
| ₹1.5–2.5k | 1.5% | 0.4% |
| ₹2.5–3.5k | 5.0% | 1.7% |
| ₹3.5–5k | 15.4% | 7.3% |
| ₹5–7.5k | 28.9% | 19.7% |
| ₹7.5–10k | 19.3% | 18.4% |
| ₹10–15k | 17.8% | 23.6% |
| ₹15–25k | 9.2% | 18.7% |
| ₹25–50k | 2.4% | 8.7% |
| ≥ ₹50k | 0.2% | 1.5% |
Two empirical regularities are visible in this raw view alone:
- The distribution peaks at ₹5–7.5k MPCE, not at the bottom — India’s “pyramid” is empirically a diamond.
- The top ~12% of Indians (MPCE ≥ ₹15k) consume ~29% of total household expenditure. The rural-urban handover happens around ₹10k MPCE.
3.3 The savings identity
Section titled “3.3 The savings identity”The accounting identity at the household level is:
Income = Consumption + Savings ⇒ Income = Consumption / (1 − s)
where s = Savings / Income is the household savings rate. Each household’s s is unobserved, so we use a schedule: s as a function of where the household sits in the population-weighted MPCE distribution. The schedule is the load-bearing assumption.
We rank every household on the people-weighted MPCE distribution — a 6-person household counts six times in the ranking — which is equivalent to ranking by MPCE in the population of individuals. Each household then receives a savings rate by linear interpolation of an anchored schedule.
3.4 Three savings schedules
Section titled “3.4 Three savings schedules”CMIE-CPHS-anchored. Anchored to the two public data points from CMIE-CPHS 2022-23 (Mahesh Vyas via PARI): poorest decile mean PCM income ₹849 vs HCES PCM consumption ₹1,805 → bottom dis-saves heavily; top decile ₹20,599 vs ₹14,993 → saves ~25-30%. Aggregate ~16% — lower than NAS (a known CMIE feature, per Pais 2021’s documentation of CMIE’s under-coverage of both tails).
PRICE/RBI-NAS-anchored. Calibrated so the aggregate household savings rate hits the RBI/NAS figure of ~22%. Top decile saves 30-40%, bottom dis-saves modestly. This is the schedule that best matches PRICE-ICE 360°’s headline that ~31% of households are middle class (HH income ₹5–30L), once household-size and people-vs-household corrections are applied.
WIL-anchored. Calibrated to the Bharti-Chancel-Piketty-Somanchi (2024) finding that the top decile captures 58% of national income and the top 1% captures 22.6%. Steepest top-end savings (top decile saves 38-48%); aggregate ~30%.
Schedules in tabular form (savings rate at percentile p):
| Percentile | CMIE | PRICE | WIL |
|---|---|---|---|
| 0 | −0.55 | −0.20 | −0.15 |
| 10 | −0.40 | −0.10 | −0.07 |
| 30 | 0.00 | 0.00 | 0.02 |
| 50 | 0.07 | 0.08 | 0.09 |
| 70 | 0.15 | 0.18 | 0.18 |
| 90 | 0.25 | 0.33 | 0.38 |
| 95 | 0.30 | 0.40 | 0.48 |
| 99 | 0.35 | 0.48 | 0.58 |
| 100 | 0.42 | 0.55 | 0.65 |
3.5 Mid-point estimator
Section titled “3.5 Mid-point estimator”For each household, project three incomes (y_cmie, y_price, y_wil) and take their simple arithmetic mean as the mid-point income y_mid. Then bin households by y_mid into PRICE-aligned annual-income tiers:
| Tier | HH annual income | PRICE label |
|---|---|---|
| Destitute | < ₹1.25L | Destitute |
| Lower Aspirer | ₹1.25–2.5L | (lower Aspirer) |
| Upper Aspirer | ₹2.5–4L | (upper Aspirer) |
| Top Aspirer | ₹4–5L | (top Aspirer) |
| Seeker | ₹5–15L | Middle class |
| Striver | ₹15–30L | Middle class |
| Rich | > ₹30L | Rich |
The ≤₹4L cutoff aligns with the upper bound of Upper Aspirer. PRICE’s official Aspirer upper bound is ₹5L; using ₹4L is a slightly tighter, more conservative line for “poor.”
4. Results
Section titled “4. Results”4.1 Mid-point distribution (headline)
Section titled “4.1 Mid-point distribution (headline)”| Tier | % of population | Million people |
|---|---|---|
| Destitute (<₹1.25L) | 2.2% | 27 |
| Lower Aspirer (₹1.25–2.5L) | 13.2% | 160 |
| Upper Aspirer (₹2.5–4L) | 25.3% | 308 |
| Subtotal ≤ ₹4 LPA | 40.5% | 494 |
| Top Aspirer (₹4–5L) | 13.0% | 158 |
| Seeker (₹5–15L) | 41.7% | 508 |
| Striver (₹15–30L) | 4.2% | 51 |
| Rich (>₹30L) | 0.7% | 8 |
Total weighted population: ~1,218M (1.22 billion).
4.2 Triangulation bounds
Section titled “4.2 Triangulation bounds”| Estimate | CMIE | PRICE | WIL | Mid-point |
|---|---|---|---|---|
| ≤ ₹4 LPA (%) | 43.0% | 39.8% | 38.7% | 40.5% |
| ≤ ₹4 LPA (M) | 523 | 485 | 471 | 494 |
| ₹1.25–4 LPA “poor-not-destitute” (%) | 39.7% | 38.0% | 37.1% | 38.5% |
| ₹1.25–4 LPA (M) | 483 | 463 | 451 | 468 |
| Aggregate savings rate | 16% | 24% | 30% | 24% |
| Gini (income) | 0.42 | 0.44 | 0.47 | 0.445 |
The three schedules disagree by only 4 percentage points on the ≤₹4L share — a tighter band than the bottom-line uncertainty would suggest, because the schedules diverge mainly at the very top (top 1%), which has negligible effect on the ≤₹4L count.
4.3 The ₹8 LPA EWS cutoff
Section titled “4.3 The ₹8 LPA EWS cutoff”The 103rd Constitutional Amendment Act (2019) introduced 10% reservation in central government jobs and education for Economically Weaker Sections (EWS), with eligibility set at household annual income < ₹8 lakh (plus asset restrictions on land and dwelling size). The income test alone — applied to our mid-point income — yields:
| Schedule | ≤ ₹8 LPA population share | Million people |
|---|---|---|
| CMIE | 81.8% | 997 |
| PRICE | 79.1% | 963 |
| WIL | 78.0% | 950 |
| Mid-point | 79.4% | 967 |
The schedules agree even more tightly here than at ₹4L — the ≤₹8L range spans just 4 percentage points. The incremental “EWS-only” band (₹4–8 LPA) captures another 473M people / 39% of the population — these are households who are not “poor” by our ₹4L cutoff but still EWS-eligible by the income test.
| Tier | % of pop (mid-point) | Million people |
|---|---|---|
| ≤ ₹4 LPA (“poor”) | 40.5% | 494 |
| ₹4–8 LPA (“EWS-only”) | 38.9% | 473 |
| ≤ ₹8 LPA (EWS income cutoff) | 79.4% | 967 |
| > ₹8 LPA | 20.6% | 251 |
Note: this is an upper bound on EWS-eligible population since EWS also requires (i) family does not own ≥ 5 acres of agricultural land, (ii) family does not own a residential flat ≥ 1,000 sqft, (iii) ≤ 100 sqft residential plot in notified municipalities, (iv) ≤ 200 sqft residential plot in other areas. Applying these asset tests would reduce the count below 967M but they cannot be cleanly computed from HCES (the survey does not record plot size or apartment square-footage). The land-area cap can be partially applied via L3’s land_area_acres.
By state, the ₹8L bound effectively covers everyone in the poorer states (≥90% in Bihar, Jharkhand, Assam, UP, MP, Odisha) and ~50-70% even in southern states. The contrast with the ₹4L bound shows how the EWS cutoff captures the entire lower-middle slab in addition to the poor:
| State | ≤ ₹4L share | ≤ ₹8L share | Million ≤ ₹8L |
|---|---|---|---|
| Uttar Pradesh | 62% | 92% | 185 |
| Bihar | 61% | 94% | 103 |
| Maharashtra | 34% | 72% | 79 |
| West Bengal | 39% | 81% | 76 |
| Madhya Pradesh | 51% | 89% | 65 |
| Rajasthan | 28% | 76% | 54 |
| Tamil Nadu | 27% | 69% | 49 |
| Karnataka | 31% | 78% | 49 |
| Odisha | 51% | 90% | 38 |
| Gujarat | 19% | 64% | 37 |
| Andhra Pradesh | 27% | 76% | 33 |
| Jharkhand | 71% | 94% | 32 |
| Assam | 53% | 93% | 30 |
| Telangana | 17% | 54% | 20 |
| Kerala | 7% | 47% | 16 |
4.4 Geographic concentration
Section titled “4.4 Geographic concentration”Sector (mid-point): 49% of rural India lives in ≤₹4L households; 19% of urban India. ~85% of the segment is rural.
State (mid-point, top 15 by population):
| State | Total pop (M) | ≤ ₹4L share | ≤ ₹4L people (M) |
|---|---|---|---|
| Uttar Pradesh | 202 | 62% | 124 |
| Bihar | 109 | 61% | 66 |
| Maharashtra | 110 | 34% | 37 |
| Madhya Pradesh | 73 | 51% | 37 |
| West Bengal | 94 | 39% | 37 |
| Jharkhand | 34 | 71% | 24 |
| Odisha | 42 | 51% | 22 |
| Rajasthan | 71 | 28% | 20 |
| Tamil Nadu | 71 | 27% | 19 |
| Karnataka | 62 | 31% | 19 |
| Assam | 32 | 53% | 17 |
| Andhra Pradesh | 44 | 27% | 12 |
| Gujarat | 58 | 19% | 11 |
| Telangana | 36 | 17% | 6 |
| Kerala | 35 | 7% | 3 |
Two stories show up here:
- Total numbers: UP and Bihar alone account for ~190M (≈ 38% of the entire ≤₹4L segment).
- Intensity: Jharkhand (71%), UP (62%), Bihar (61%) are the states where the median household sits below ₹4 LPA.
The “BIMARU + East” cluster (UP, Bihar, MP, Jharkhand, Odisha, WB, Assam) holds 70% of the segment.
5. Caveats and Limitations
Section titled “5. Caveats and Limitations”-
HCES self-declared MPCE vs imputed MPCE. This analysis uses the
MONTHLY_CONSUMPTION_EXPfield from L15, which is the respondent’s self-declared total at the end of each visit’s questionnaire. MoSPI’s published headline MPCE for the same survey is rebuilt from the item-level files (L5/L9/L14) with imputation for free PDS items, in-kind transfers, and durables annualization — and runs roughly 50–65% of the self-declared value. Using imputed MPCE would shift every household leftward and push the ≤₹4L share to 55-65% of the population (~670-790M), much closer to PRICE’s combined Destitutes + Aspirers headline. The figures here are best read as a robust lower bound; a fully MoSPI-comparable estimate would require re-aggregating from L5/L9/L14. -
The savings schedules are interpolations of public anchor points, not directly measured curves. CMIE-CPHS does not publish a decile savings table; we have only the D1 and D10 anchors. PRICE and WIL similarly publish aggregates rather than per-decile savings curves. The shapes between anchors are reasonable interpolations but not directly verified.
-
Sample frame limits at the top. HCES, like all consumption surveys, under-represents the very rich. Even the WIL schedule does not fully recover ultra-high incomes since the underlying HCES sample lacks billionaire households. The bottom end of the distribution is comparatively well-covered.
-
Household size assumed constant per household. Income/consumption are HH totals; the mapping to “population in band” uses HH size at the time of survey. Multi-generational households can blur the income/consumption identity if some members receive in-kind transfers.
-
No life-cycle adjustment. A young household consuming more than it earns (e.g., student-led) and a retired household drawing on assets both register as “dis-savers” in this framework. Cross-sectional savings rates are mechanically influenced by age structure.
-
Per-capita-MPCE-percentile applied to HH-level consumption. The savings rate schedule indexes on where the household falls in the per-capita distribution but is applied to its absolute (HH-level) consumption. This is the standard approach but implicitly assumes a HH’s “richness ranking” depends on per-capita living standards, not aggregate spend.
6. Reproducibility
Section titled “6. Reproducibility”All artifacts live in this repository.
Scripts (run in this order):
| Script | Purpose | Outputs |
|---|---|---|
transform_hces.py |
Builds the one-row-per-household master from L1, L3, L15 | hces_household_master.csv / .parquet, dictionary |
consumption_pyramid.py |
Empirical consumption distribution (§3.2) | consumption_pyramid.png, consumption_pyramid_table.csv |
estimate_income.py |
Single-schedule income projection (CMIE or RBI-anchored) | hces_household_master_with_income.csv, income_vs_consumption.png |
compare_income_views.py |
Side-by-side pyramids under all three schedules | income_views_compare.png, .csv |
estimate_poor_population.py |
Triangulated ≤₹4L analysis with state breakdown | poor_population_below_4LPA.png, poor_population_by_state.csv, poor_population_by_band_*.csv |
midpoint_estimate.py |
Mid-point estimator (this paper’s central result) | midpoint_pyramid.png, midpoint_estimate.csv, midpoint_estimate_by_state.csv |
ews_cutoff.py |
≤₹4L and ≤₹8L cutoffs with EWS context | ews_cutoff_pyramid.png, ews_cutoff_summary_*.csv, ews_cutoff_by_state.csv |
Key derived columns (in hces_household_with_income_all_schedules.parquet):
| Column | Meaning |
|---|---|
mpce |
Monthly per-capita expenditure (₹) |
y_cmie / y_price / y_wil |
Projected HH annual income under each schedule (₹ lakh) |
y_midpoint |
Simple mean of the three (₹ lakh) — this paper’s central income measure |
weight |
Survey multiplier ÷ 100 |
people_weight |
weight × hh_size (use this for population aggregations) |
7. References
Section titled “7. References”- NSS Household Consumption Expenditure Survey 2023-24 microdata — MoSPI, December 2024. Source page
- Mahesh Vyas / CMIE — The Distribution of Household Income, 2019-2024 (via People’s Archive of Rural India). Link download
- Pais, J. (2021) — CMIE’s Consumer Pyramids Household Surveys: An Assessment (Society for Social and Economic Research). PDF — documents CMIE’s under-coverage of both tails and the implied bias in aggregate savings. download
- PRICE / Rajesh Shukla — The Rise of India’s Middle Class: Results from PRICE’s ICE 360° Surveys (2022). PDF and brochure. download 1, download 2
- Bharti, N. K., Chancel, L., Piketty, T., Somanchi, A. (2024) — Income and Wealth Inequality in India, 1922-2023: The Rise of the Billionaire Raj, World Inequality Lab Working Paper 2024/09. PDF download
- The India Forum — CMIE’s Consumer Pyramids Household Surveys: An Assessment (review). Link download
- RBI Handbook of Statistics on the Indian Economy — for the household savings ratio in National Accounts. RBI
Document version 1.0. Microdata transformations and figures generated 2026-05-14.