The impact of survey design on global and regional poverty trends

In low- and middle-income countries, monetary poverty is measured using data from household surveys that elicit the income or consumption patterns of a representative sample of households in a country. Numerous studies have shown that the design of household surveys impacts the data collected, measured income or consumption, and therefore also measured poverty rates.

For example, when measuring food consumption, it matters whether households are asked to keep a diary of their consumption or to recall their consumption. In the latter case, it is of consequence whether they are asked to recall their consumption over the past seven days or 30 days. For both food and non-food consumption, it matters how disaggregated the elicited categories of consumption are, for example whether households are asked to report their rice consumption in aggregate, or brown and white rice separately.

To ensure that household surveys are on the frontier of poverty measurement, data collection practices and questionnaire design occasionally need to be revisited. Though this creates more accurate poverty estimates, it also results in comparability breaks in estimated poverty rates. At times, measured consumption increases by over 50% because of changes to the survey design and construction of consumption aggregates, which highly biases poverty trends using fixed poverty lines if unaccounted for. 

For single countries, solutions to this problem are well known and frequently applied, and include survey-to-survey imputations that predict the distribution of consumption had there been no change to the questionnaire design or bridge surveys that collect an old consumption aggregate on a subsample in a newer survey. Yet the problem extends beyond single countries. To understand how poverty is evolving globally and by regions, such methods cannot be applied consistently. As a result, the problem is most often ignored globally.

In this blog (and associated paper), we use growth from national accounts to bridge non-comparable sequences. That is, if consumption grew by 50% between two non-comparable estimates but economic growth from national accounts was substantially less, we will use the latter to adjust past consumption aggregates. We apply this method to the World Bank’s global poverty numbers in the Poverty and Inequality Platform (PIP) to shed light on whether the trends in poverty are robust to differences in how consumption is measured within countries over time.

After 2013, the global extreme poverty rate in PIP differs only marginally from our series that deals with comparable breaks. In 2013, PIP’s series declines 2.4 percentage points while the comparable series declines by only 1.0 percentage point. This coincides with a comparability break in China, which added included rent and made more changes to its measured consumption in 2013. This increased measured consumption and hence lowered poverty in China by 31 million people. The comparable series attempts to mitigate this break.

In the 1990s and 2000s, the trends are relatively parallel but systematically a little lower with the comparable series than in PIP, largely reflecting the comparability break in in China in 2013. The next large change in trends happens from 1988 to 1989, where PIP has an increase in poverty of 1.5 pct. points, while the comparable series declines by 0.8 pct. points. This again reflects an incomparability in China, which went from an income to a consumption aggregate from 1987 to 1990. These two breaks together mean that the overall decline in poverty from 1981 to 2024 is broadly similar.

This masks some larger differences at the regional level. The divergences in the trend for East Asia & Pacific largely reflects the global pattern and the comparability breaks for China. In Europe & Central Asia, the trends are rather well aligned except for a large difference in 2012 when Uzbekistan switches from a consumption to an income aggregate. In South Asia, the comparable series aligns closely with the series in PIP in recent decades. This alignment is primarily due to the resolution of a significant comparability break in India when it transitioned from a uniform to a mixed modified reference period, which has been addressed at the country level in PIP and is thus incorporated into both series. In Sub-Saharan Africa, the comparable trend from 2014-2019 differs noticeably from PIP, with PIP suggesting a small decline in poverty but the comparable series suggesting a small increase. Together with the COVID-induced increase in poverty in 2020 and the stagnation in the years that followed, poverty is higher in Sub-Saharan Africa in 2024 than in 2014 using the comparable series but not in PIP. 

At the country-level, the changes are often much starker. All country level changes as well as two alternative ways of creating comparable series are available in the background paper.

The authors gratefully acknowledge financial support from the UK government through the Data and Evidence for Tackling Extreme Poverty (DEEP) Research Programme.