The Poverty Line Paradox

How five methodological choices shape who counts as poor

When the World Bank reports a poverty rate, the number appears as a fact about the world. This analysis examines how five methodological choices — each invisible to the headline number — shape poverty statistics as much as the underlying economic reality.

240×
range of "poor Indians" depending on poverty line and PPP vintage (10M to 969M)
75%
of annual poverty data points are modeled, not measured by household surveys
26
survey comparability breaks across 10 countries — each fractures the trend line

1. The poverty line multiplier

The international extreme poverty line of $2.15/day is a methodological choice, not a natural boundary. Raising the threshold to $3.65 or $6.85 multiplies the measured poverty headcount dramatically — from the same underlying data.

Same Data, Different Poverty Lines
Headcount ratio at three thresholds · most recent survey year, 10 countries
$2.15/day
$3.65/day
$6.85/day
0%25%50%75%100%
CountryYear$2.15$3.65$6.85Ratio

Across these 10 countries, moving from $2.15 to $6.85 multiplied the measured poverty headcount by an average of 12×. Indonesia's ratio is 44×: almost no one is below $2.15, but a majority is below $6.85. The poverty line isn't measuring different degrees of the same thing — it's defining different populations entirely.

2. The India summary

India condenses all five measurement layers into a single case.

India 2022: One Country, Four Numbers
Poverty headcount by threshold · same survey, same year
0%20%40%60%80%
ThresholdHeadcountPeople (est.)
$1.90/day (2011 PPPs)0.3%~4M
$2.15/day (2017 PPPs)0.8%~10M
$3.65/day (2017 PPPs)13.8%~193M
$6.85/day (2017 PPPs)69.2%~969M

The same country, same year, same underlying survey: the number of "poor" Indians ranges from 4 million to 969 million depending on the poverty line and PPP vintage. That's a 240× range. India's poverty story is either "nearly eliminated" or "the majority" — determined entirely by the measurement layer.

India's 2022 estimate also uses a new survey instrument (HCES) that is not comparable with the previous one (NSS-SCH2, last used in 2011). The apparent trajectory from 6.6% to 0.8% crosses a comparability break. The headline "India cut poverty from 7% to 1%" draws a line through two incompatible measurement systems.

3. Three-quarters of data is modeled

The World Bank provides annual poverty estimates for most countries. But most of these are not measured — they are modeled through interpolation between actual household surveys, using GDP growth rates and assumed pass-through elasticities.

Actual Surveys vs Interpolated Data Points
Share of annual data points (2000–2025) based on actual household surveys
Actual surveys
Interpolated
CountryTotal pointsActual surveysInterpolatedSurvey %

Of 270 annual data points across 10 countries, only 68 (25%) are based on actual household surveys. The DRC has actual surveys for just 11% of its filled data points. The global poverty aggregates that track SDG progress are built primarily from modeled data.

4. PPP rebasing shifts everything

The international poverty line is denominated in Purchasing Power Parity dollars. When the ICP updates PPP conversion factors — as it did in 2005, 2011, and 2017 — all poverty statistics shift. The $2.15 line (2017 PPPs) and the $1.90 line (2011 PPPs) were designed to be "equivalent," but they produce systematically different answers.

The PPP Vintage Effect
Headcount difference: $2.15 (2017 PPPs) minus $1.90 (2011 PPPs) · percentage points
0pp2pp4pp6pp8pp
CountryYear$2.15 (2017)$1.90 (2011)Difference

The average shift is 4.1 percentage points. In Tanzania the gap reaches 8 points — 11 million additional people counted as extremely poor purely through a PPP rebasing. The direction is systematic: 2017 PPPs produce higher poverty estimates across virtually every country.

5. Survey breaks fracture trend lines

Poverty trends require comparable surveys over time. But survey methodology changes. The PIP data explicitly flags these through "comparable spells" — periods within which surveys are methodologically consistent.

Survey Comparability Breaks
Number of distinct comparable spells per country · each break means the trend line fractures
02468
CountrySpellsBreaksSurvey instrumentsWelfare type

Across 10 countries, there are 26 comparability breaks — an average of 3.6 comparable spells per country. India has 5 breaks across 4 different survey instruments. Indonesia has 6 breaks despite using a single instrument, because revisions within the same survey family create discontinuities. When a headline reads "poverty fell from X% to Y% over three decades," it is drawing a line through multiple incompatible measurement systems.

The measurement layer pattern

This is the third data domain — after IATI aid transparency data and governance indicators — to exhibit the same structural pattern.

Three Domains, One Pattern
How the measurement layer shapes the headline number
Domain The headline The measurement layer
IATI "85 orgs in Uganda's governance sector" 21% are measurement artifacts
WGI "India ranks 70th in governance" 90% CI spans rank 31–132
Poverty "0.8% of Indians are poor" Range is 0.3%–69.2% by method

In each case, a specific number is presented as a fact about the world. Methodological choices invisible to the end user shape the number as much as the underlying reality. The choices are defensible — they are not errors — but they are not disclosed alongside the headline. Consumers treat the number as more precise than it is.

Poverty statistics are not broken. The World Bank's methodology is transparent to specialists. The PIP API exposes all the metadata needed to understand how each number was produced. The problem is the gap between production and consumption: poverty numbers are produced with explicit methodological caveats but consumed as simple facts.

This is not a call to distrust poverty data. It is an argument that the measurement layer should travel with the number. A poverty headline without its methodology is not a fact — it is a fragment of a fact.

Method

All data retrieved from the World Bank Poverty and Inequality Platform (PIP) API. Poverty headcounts compared across three international poverty lines ($2.15, $3.65, $6.85/day in 2017 PPPs) and two PPP vintages (2011, 2017). Interpolation analysis compares gap-filled vs survey-only data points for 10 countries, 2000–2025. Survey comparability assessed using the PIP's own comparable_spell metadata. Analysis code: research/poverty_measurement.py.