Bottom coding replaces extremely low reported income or consumption values to improve the accuracy of global poverty and inequality estimates. / Image: Shutterstock
The World Bank’s Poverty and Inequality Platform (PIP) started bottom-coding the income and consumption distributions that underpin its poverty and inequality statistics. Bottom-coding means replacing all values below a threshold with the threshold. Following Yonzan ⓡ al. (2026), the threshold is set at 25 cents per person per day in 2017 PPPs and 28 cents in 2021 PPPs. Bottom-coding is needed to avoid a situation where a very small number of extremely small observations, that could simply reflect measurement error, disproportionately affect countries’ rankings on distribution-sensitive welfare measures. This blog outlines the need for a bottom code and shows that the impact on the reported estimates is very small. Finally, it also explains how the bottom code will be updated with any future revisions in the PPPs.
The need for a bottom code
In many cases, temporary conditions faced by some households, or simply errors in data collection, mean that households may report improbably small values for income or consumption. For example, some households in rich countries report income less than 1 cent per day, and some households in other parts of the world report consuming very little. Zero income is possible for a limited period of time, such as during periods of unemployment when a household might live off savings, but it is unlikely to reflect the long-run financial situation of the household. Very low values of reported consumption are more problematic and cannot be reconciled with the fact that across countries it costs at least 24 cents a day (in 2017 PPPs) to eat a diet that provides sufficient amount of energy to survive – such as a diet consisting only of basic staples like potatoes.
These measurement concerns can affect the level and cross-country ranking of poverty and inequality. To give an example, Figure 1 reports the ranking across countries using the Prosperity Gap measure. The Prosperity Gap shows how far a country is from a global standard of prosperity while giving much greater weight to individuals that are far away from the line (for detail, see here and here). When using the raw data, wealthy countries like Austria, Norway, and Sweden would be classified as less prosperous (a higher Prosperity Gap) than Peru, whose average income is 6 times lower. This is because these countries have a small share of the population that report extremely low incomes.
Figure 1. Rich countries can be ranked as less prosperous due to measurement error
All summary measures that are sensitive to the income or consumption of the poor respond strongly to extremely small values, as they should. Many measures also cannot incorporate zero and negative incomes. For both these reasons, researchers and data providers commonly use bottom codes. Historically, PIP had recoded zero values as 1 in the computation of the Mean Log Deviation, a bottom-sensitive inequality measure which cannot accommodate zeros. Since 2024, PIP systematically bottom-codes all distributions from which all poverty and inequality statistics are then estimated. Specifically, any consumption or income value between 0 and 25 cents (inclusive) is replaced with 25 cents (in 2017 PPPs). As was done previously, observations that are negative are excluded entirely from the analysis as there is evidence that those individuals may not be poor but could be business-owners with an income shock in the period that data was collected. The threshold is set by triangulating various statistical methods based on over 800 consumption surveys and the biological consumption minimums.
Impact of the bottom code on poverty and inequality estimates
With the bottom code, Austria is ranked 14th richest out of 46 (as shown in Figure 1) rather than 42nd without the bottom code. Likewise, Norway is correctly ranked as 6th not 41st and Sweden is ranked 16th not 44th. Peru is correctly ranked 42nd with the bottom code compared to 36th without.
Of the almost 2000 household surveys that underpin PIP in the past 25 years, close to 95 percent have less than 1 percent of observations with observations below 25 cents (Figure 2). For these surveys, the impact of bottom-coding is marginal. Only 17 surveys have more than 2 percent of the population reporting below 25 cents.
Figure 2. Share of population living below 25 cents
A greater number of pre-2000 income surveys reported zero or negative incomes, but even in such cases, bottom-coding does not affect headcount poverty measured at extreme thresholds (i.e., 2.15 in 2017 PPPs) or higher and has practically no effect on average income or consumption. There are small changes for distribution sensitive poverty measures such as the squared poverty gap. The most extreme impact for a survey in the last 25 years is a 2 percent reduction in the squared poverty gap with the bottom code in Lesotho’s 2002 survey (from 24 to 23.5). The impact on estimates of inequality depends on how sensitive the measure is to changes at the bottom tail of the distribution. The Gini (an inequality measure that’s sensitive around the middle) in recent years is impacted at most 1 percent (again for Lesotho 2002). The largest change in the mean log deviation is a 6 percent decrease (0.51 to 0.48) for the same country.
Updating the bottom code with different PPP rounds
The World Bank began using the 2021 PPPs (Purchasing Power Parities) to calculate global poverty and inequality in September 2024. The 25 cent threshold derived using the 2017 PPPs was updated to 2021 PPPs using the average price change across countries, called the delta ratio. The delta ratio compares prices between two PPP rounds. For the US, it is the domestic inflation rate between those years, but for other countries it also captures both changes in PPPs and domestic inflation. Between 2017 and 2021 PPPs, the delta ratio is 1.13. That means the bottom code changes from 25 cents in 2017 PPP to 28 cents in 2021 PPPs (25 x 1.13). Earlier, the average delta between 2011 and 2017 PPPs was also 1.13, so the bottom code in 2011 PPPs is 22 cents (22 ÷ 1.13).
Would you like to be updated with the latest news on PIP? Register to our newsletter here.
The authors gratefully acknowledge financial support from the UK Government through the Data and Evidence for Tackling Extreme Poverty (DEEP) Research Program.
Join the Conversation