11/23/2022 | Press release | Distributed by Public on 11/23/2022 07:57
The World Bank has continuously tracked global poverty at least since 1990 (for example, see World Bank 1990; Chen and Ravallion, Chen 2010; Ferreira et al. 2016, World Bank 2022). This also coincides with the beginning of our contemporary period of widely available household survey data which are needed to measure poverty. There is, however, a void for many countries before the 1990s. Household survey data is far more scarce in the 1980s and almost non-existent before then. Historical global poverty estimates have been offered by Bourguignon and Morrison (2002) and an OECD 2021 report. We outline a different method to extend the data series of annual global poverty to the end of the Second World War. Our main motivation for estimating historical poverty rates is to contextualize the size of the COVID-19 poverty shock in a broader historical perspective (see World Bank 2022 for a recent discussion).
The World Bank's Poverty and Inequality Platform (PIP) hosts country-level household survey data starting around the 1980s. With each decade, there has been an increase in the number of countries with an available household survey. Fewer than 16 countries reported household survey data in a typical year in the 1980s compared to 37 countries in the 1990s and more than 75 countries in the following two decades (Figure 1).
One criterion selected by the World Bank for reporting global poverty is that the available survey data covers at least 50% of the population of low and lower-middle-income countries (within a three-year window), as well as the global population overall. This criterion is met for the years starting in 1981 and now up to 2019. However, there are reasons to scrutinize the data further in the 1980s. While the 50% minimum population threshold for the lower-income countries is met in the 1980s, this is driven mostly by available data in China, which has survey data for the years 1981, 1983, and 1987. For instance, 40% of the global extreme poor in 1987 lived in China, a low-income country in that period. At the same time, survey coverage in Sub-Saharan Africa was below the 50% threshold in the 1980s, so no regional aggregate is reported in PIP.
Additional uncertainty is introduced as there is a switch in the welfare aggregate used for China-income was used to measure welfare until 1987, while subsequent datapoints (in 1990 and beyond) use consumption instead. Conceptually, consumption is less than or equal to income on average-income can be used for consumption or savings; and so switching from one measure to the other will create a break in the series. Given the size of China in the global poverty counts during this era, this switch not only is a break in comparability in the Chinese poverty trends but also a break in comparability for the global poverty estimate. Between 1988 and 1989, global poverty (at the $2.15 line in 2017 PPP dollars) increased by 170 million people with China contributing 154 million. While we do not have survey microdata for China in those years, national accounts-based data suggests that between 1988 and 1989 household incomes grew not fell. Therefore, this increase in poverty is largely an artifact of the shift in the choice of the welfare aggregate in China from income in the 1987 survey (with poverty estimated at 67.5%) to consumption in the 1990 survey (poverty estimated at 72%). To obtain annual estimates of global poverty, these datapoints are interpolated (as described in Prydz et al. 2019)
Given the combination of relatively limited survey data in most countries in the 1980s, the break in comparability for China pre- and post-1990, and China's influence on the changes in global poverty in this period, we "backcast" global poverty prior to 1990 rather than using the available microdata before 1990 for this exercise.
Starting with distributions of income or consumption for each country from PIP, all lined-up to 1990, we extrapolate each distribution backward using growth rates from PIP (largely based on WDI) supplemented by growth rates using Bolt and van Zanden (2020). We are able to do this for 164 World Bank economies. In the years when no growth data is available for these countries (mostly because the country did not exist in its present form), we assign the average growth of the region using the regional definitions in PIP. For the remaining countries with no microdata in PIP, we assign an average welfare distribution of the region. This gives us a distribution for each of the 218 World Bank economies for the period 1950 to 1989 similar to what we have in the period after 1990. We refer to the poverty estimates derived using this method as backcast estimates.
Note that the extrapolations are based on the estimates of growth rates of per capita GDP for each country. These growth rates have no information about the dispersion in the welfare changes within each country. Hence, our backcast data assumes distribution neutrality within a country for the entire period from 1950 to 1989. While changes in inequality certainly affect poverty, historically, most changes to poverty have been driven by growth rather than distributional changes. Additionally, there is evidence that within-country inequality stayed relatively stable in this period which gives additional confidence to our estimate.
We find that there was a gradual decline in global extreme poverty-that is, those living below the $2.15 poverty threshold in 2017 PPP-in the period from 1950 to 1990. Global poverty declined by about twenty percentage points over those 40 years. Figure 2 splices together three series: the backcast series from 1950 to 1990, the PIP data from 1990 to 2019 (labeled as historical data) and projected estimates for 2020 from the work by Mahler ⓡ Yonzan ⓡ Lakner (2022) (also see World Bank 2022).
Figure 3 shows the same series, but focuses on the annual changes. We find that global extreme poverty increased at only three points in time during the backcast period-in the mid-1960s, in the early 1970s, and the late 1970s. The first episode is mostly explained by rising poverty in India, which experienced a tumultuous economic period in the mid-1960s (a drought, the war with Pakistan, devaluation). The latter periods roughly coincide with the first and second oil crises. For every other year, we observe (at least modest) declines in global poverty.
On average, poverty declined by 0.5 percentage points annually from 1950 to 1990. This rate of poverty reduction then doubled to 1 percentage point annually in the period after 1990 leading up to 2019. For the entire period after World War II up to 2019, global extreme poverty increased only in four episodes-the first three discussed above, and the last during the mid-1990s triggered by the Asian financial crisis. Using these estimates, the shock due to COVID-19 is the largest negative shock to global poverty since WWII.
The OECD's publication "How was life?" also provides long-run estimates of extreme poverty. The largest estimated change in the OECD data series is a 1 percentage point increase in poverty rate from 50 to 51 percent in 1954. It is not clear what event drives this increase, but if it is real, it is a slightly larger percentage point increase than the 0.9 percentage point increase recorded in 2020 in our data. However, it was a much smaller increase in the number of poor people (in both absolute and relative terms) than recorded in 2020. We also looked at significant global events or events in large countries such as China and India that have occurred since the Second World War, such as the 1947 partition of British India, that may have increased poverty. None of these events was global enough in reach to result in the size of poverty increase seen in 2020. Perhaps the most significant global event considered, The Great Leap Forward in China (1958-1962), was a severe and large income shock, but it did not result in a large increase in global extreme poverty. This is because at that moment in time extreme poverty in China is estimated to have already been higher than 90 percent and thus had little scope to increase further.
A lot of uncertainty surrounds the estimates of global poverty. A large part of this uncertainty stems from the unavailability of household survey data (see here and here). As discussed above, our backcasting method ignores the changes in inequality between households within a country. In addition, there are uncertainties in countries' GDP estimates themselves. Studies (see here and here) have highlighted the concerns in the reported GDP from recent years, and there are additional uncertainties around GDP estimates for the years before 1990. Therefore, the backcast estimates of global poverty reported in this blog should be viewed taking into consideration these gaps in knowledge. However, despite these caveats, the available evidence indicates that 2020 was a solemn and historic year for many reasons, including the unprecedented rise in global poverty.
The authors gratefully acknowledge financial support from the UK Government through the Data and Evidence for Tackling Extreme Poverty (DEEP) Research Program.