Global child mortality: it is hard to overestimate both the immensity of the tragedy, and the progress the world has made

Disastrous events with high death tolls always make the headlines, and rightfully so. Yet there are many daily, recurring tragedies in the world which create as much or more suffering and often go unnoticed.

The Institute for Health Metrics and Evaluation (IHME) estimates that in 1990 more than 12 million children died. Most of these deaths were preventable, and arose from illness and poverty.

Let’s put this in perspective. Aviation disasters are a form of tragedy that is both relatively uncommon and widely reported. A Boeing 747 can carry up to 620 people. The total number of child deaths in 1990 is equivalent to 53 massive aircraft disasters every day, involving primarily children, with no survivors.

While every plane crash makes the headlines and 53 of them in a single day would have media coverage everywhere, we see proportionally less reporting about the more prevalent causes of mortality. By considering how the unseen, daily cruelty of child death compares to discrete tragic events, we can find a sense of perspective about the scale involved.

Here at we present the empirical evidence of global development. To understand these issues and see the progress involved, we need to have a look at long-term changes over the course of history.

A ten-fold decrease in child mortality

The below chart shows the share of children surviving the first five years of their lives. It goes back to the year 1800. Health conditions in the time of our ancestors were such that more than 4 out of 10 newborns died before their fifth birthday. The historical estimates further suggest that it was the entire world that lived in poor conditions: there was relatively little variation between world regions, and even in the best-off countries of 1800 every third child perished.

Children regularly dying was a reality for all families for a very long time. For millennia our ancestors lived in poor health and most had short lives.

The chart shows that this finally began to improve in the 20th century, as humanity achieved broad improvements in health and living conditions. In the first half of the 1900s, the global child mortality rate fell by more than half and yet was still high: every fifth child born in 1960 died before 1965. Fortunately, in recent decades – in our lifetimes – we have seen rapid progress. In 2015 global child mortality fell to 4.3% – 10-fold lower than two centuries ago.

One reason we do not see this progress is that we are unaware of how much worse the past was. A long-term perspective is crucial.

Child mortality declined in all countries

The decline in global child mortality is not just due to improvements in a few well-off, developed regions. In fact, every single country achieved a reduction. This is what the chart below shows. Hover over the lines with your mouse to see the decline of child mortality in each country.

Yet, even after decades of unprecedented global progress there are still stark divides. The difference between the worst-off countries (e.g. Angola with a 16% child mortality) and the best-off countries (e.g. Luxembourg at 0.2%), shows what is possible and underscores that a lot of work remains to be done.

This work is already underway. The biggest recent improvements were achieved in the countries that were worst off in the 1950s. The least healthy countries have been catching up, and the inequality in health across the world today is substantially lower than a few decades ago.

With fewer children dying, we might initially expect populations to increase, leading to problems with living space and resource consumption. Thankfully, in the long term this does not appear to be the case: as countries develop and mortality falls, they also undergo a decrease in birth rates known as the demographic transition. This trend is expected to continue to apply to those developing countries which currently have both high mortality and high birth rates. We recently made a short video about this with our colleagues at Kurzgesagt.

We are making progress against a broad range of causes of child death

What is killing our children? Here we compare the causes of child death in 1990 and in 2015. The data comes from the IHME’s Global Burden of Disease study.

This visualization shows the many causes of child deaths and the mortality rate that they are responsible for. Birth complications, pneumonia, diarrheal diseases, and malaria all still lead to the death of more than 400,000 children annually.

In many cases, the comparison with 1990 shows that we are on track to a rapid reduction. But the breakdown by cause also highlights that preventable diseases are still responsible for an unacceptably large share of child deaths and we have the duty and the possibility to reduce child mortality further.

27 major plane crashes averted every day

The achievement that this blog post highlights is that child mortality has decreased substantially. What does this mean in terms of the absolute number of child deaths that we discussed above?

The number of child deaths declined from over 12 million in 1990 to 5.8 million in 2015 (the year for which we have the latest data). Notably, this decline by 6.3 million child deaths happened despite the absolute number of births increasing slightly over the same period.

6.3 million fewer deaths means that compared with 1990, there are 17,258 fewer child deaths every single day.

We were also surprised to see the scale of the reduction of child mortality when broken down like this. Over a period of 25 years, in which the number of births had slightly increased, the world was able to reduce the number of child deaths every single day by more than 17,000. Compared with 1990, this is the equivalent of saving 27 planes from crashing on a daily basis. Were it to receive as much attention as aviation disasters do, we think this would be seen as an extremely big deal.

Every case of a family losing a child is a tragedy. And even after an impressive improvement in global health, the unseen daily cruelty of child death remains immense: close to 6 million children still die every year; 16,000 children every day.

The enormous global tragedy of preventable child death, and the progress against it, are best seen over time scales too long for a standard media cycle. We should still strive to be aware of both. The number of deaths remains very high, but the historical perspective gives us hope that a future where many more children are able to live full lives is possible.

Does the impressive historical decline in poverty capture non-market transactions?


This blog post draws on data and research thoroughly discussed in our entry on world poverty


The last two centuries have seen a remarkable improvement in living conditions around the world. The chart below shows estimates of the share of the world population living in extreme poverty. As can be seen, in 1820 the vast majority of people in the world lived in conditions that we would call extreme poverty today. Since then, the share of people living in extreme poverty has fallen continuously. This is a remarkable achievement.

Many people are not aware of this decline of extreme poverty. And among those discussing these statistics, significant confusion remains about the extent to which the figures truly reflect changes in poverty over time, rather than changes in income sources. For instance, in a discussion of the below chart on reddit, one user noted: “That is indicative of the fact that quite a lot of the world was enslaved to Colonial powers or did not use fiat currency.” In other words, the argument frequently put forward is that the decline of poverty seen in the chart is an artefact of measuring welfare by relying solely on market transactions, and therefore it fails to take into account non-monetary sources of welfare.

Here we want to briefly explain how long-run estimates for absolute extreme poverty are obtained, and focus on how non-market transactions – specifically non-monetary forms of income, such as subsistence farming – are taken into account.

Our main message is that, despite significant limitations in data quality, it is not the case that the observed progress in the reduction of poverty is an artefact of the estimation methodology.

Where do long-run poverty estimates come from?

The most straightforward way to measure poverty is to set a poverty line and to count the number of people living with incomes or consumption-levels below that poverty line. The methodology used by the World Bank to measure extreme poverty today follows this approach.

For the purpose of measuring ‘extreme poverty’, the World Bank currently uses a poverty line equivalent to $1.90 international dollars per day. The metric (‘international dollars’) can be thought of as a hypothetical currency that aims to correct for the changing value of money over time (inflation), as well as for the fact that people in different countries face different price levels. You can read more about this metric in our entry on economic growth.

The level of the ‘international poverty line’ was chosen by the World Bank in a way that reflects the national poverty standards set in the poorest countries for which national poverty lines are available. You can read more about where the poverty line comes from in our entry on world poverty.

Given the metric (international dollars) and the threshold ($1.90), measuring poverty requires one more key ingredient: per capita consumption.

For the recent past (1981 until today), household surveys are the most common source of data to estimate per capita consumption. For the more distant past, the main sources are academic studies that reconstruct historical income levels from cross-country macro estimates on economic output and inequality. Let’s examine each of these sources.

How is per capita consumption estimated from household surveys?

Consumption is defined as “the use of resources whether acquired through purchase (expenditure) or through household production or provided from outside the household, such as by relatives, charities, or the government”.

In principle, one could use household surveys to estimate (i) resource outflows (monetary expenditures, home production and transfers); (ii) resource inflows (earnings and other non-market sources of income such as, again, home production and transfers); and (iii) change in assets between the beginning and end of the relevant period (including savings, owned durable goods, etc.).

Given all this information, consumption, as per the definition above, could be estimated directly from (i), or as the difference between (ii) and (iii). Either approach would give the same result. In practice, however, surveys on expenditures are different to surveys on incomes.

For the majority of countries, the World Bank estimates consumption directly from household surveys on expenditures. For a significant minority of countries, however, the World Bank estimates are based on income surveys. Notably, in both cases, the estimation methodology does include home production and transfers, by attaching monetary values to such non-market transactions.

How are monetary values placed on things like food grown at home and gifts from relatives? One common approach is to ask survey-respondents about the amount of such resources consumed over a given reference period. The aim is to then ascribe a monetary value to the reported consumption. This is done by multiplying the consumed amounts by extrapolated market prices. A second approach asks households directly about their own valuation of the amount of money they would expect to pay if they had bought such items themselves, or, the amount of money they would expect to receive if they had sold these items. The second approach is commonly used to establish a rental equivalent for housing and durable goods owned by the household.

How are income and expenditure surveys actually conducted? Different countries use different surveying instruments, and while there is much scope for harmonization (see Beegle et al 2012), there are some basic common features that allow for cross-country comparisons. In most cases, surveys are representative at the national level and record responses provided by ‘primary respondents’ such as the head of the household. Respondents report expenditures (or incomes) either by answering questions from memory (the ‘retrospective recall method’) or by consulting written records (the ‘diary method’). In the case of expenditures, different reference periods are used to record responses across different categories of goods, with longer periods for goods or services that tend to be acquired less frequently.

Clearly, there are several issues with the quality of the estimates that are obtained from applying the various methodologies we have discussed here. And specifically in the context of non-market transactions there are certainly important limitations of the data, independently of the methodology adopted. For example, it is difficult to monetize the private benefits from consuming public goods (e.g. infrastructure such as roads but also education). Yet the fact remains: World Bank global figures on extreme poverty do aim to take into account non-market transactions.

Deaton and Zaidi (2002) provide guidelines for constructing consumption aggregates from survey data, specifically in the context of poverty measurement and welfare analysis. The following table, from their paper, provides some details regarding the main components of consumption aggregates in several countries. As can be seen, home-produced food is indeed an important form of consumption captured by household surveys.

Main components of the consumption aggregate – Deaton and Zaidi (2002)

How do researchers reconstruct historical poverty estimates?

Historical estimates of poverty come from academic studies that reconstruct income levels in the past, using cross-country macro estimates on economic output and inequality.

A seminal paper following this approach is Bourguignon and Morrison (2002) and their work is the source of the estimates for the time 1820 to 1970 shown in the first graph of this text. Bourguignon and Morrison’s starting point is to estimate the global distribution of incomes over time. The change of extreme poverty is then calculated via changes in the share of the world population with incomes below the poverty line, according to the corresponding estimated distribution of incomes.

Bourguignon and Morrison (2002) rely on three types of data in order to estimate the distributions of income: economic output (real GDP per capita), population and inequality. The first two sources provide information regarding ‘the size of the pie’, while the third one provides evidence regarding the distribution of that pie.

The approach outlined above leads to a natural question: How can researchers construct economic output for the distant past? Fouquet and Broadberry (2015) provide a detailed account of how economic historians construct these estimates. It is painstaking work with which researchers occupy themselves for years. The generally preferred approach to estimating national income is the output approach, which relies on historical records by economic sector. For example, for agricultural production, researchers use church records for the estates of farmers, as well as accounting documents produced by farmers and kept in local record offices. Agricultural outputs are then calculated by multiplying the acreage for each crop by the yield per acre. Once this is established, prices for individual crops and animal products are used to convert the output into current prices and create weights for an ‘agricultural real output index’. Outputs related to other sectors, such as leather and food processing, are estimated using a similar approach applied for the specifics of each sector. Finally all these series are then brought together using a set of sectoral weights that capture the changing structure of the economy.

Once again, the estimates obtained from applying this methodology are surrounded by a considerable margin of error; but the point we want to emphasize is that those estimates of poverty do take into account non-market transactions such as subsistence farming.

How are macro and micro estimates of consumption related?

We have already noted that long-run series on poverty combine information from national accounts (macro data) and household surveys (micro data). What is the conceptual relationship between these two sources? This question is particularly relevant for the period after 1981, when both sources become simultaneously available.

The following table, from Atkinson (2016), provides some definitions for household-survey (HS) and national-accounts (NA) concepts of consumption.

The main message from this table – in the context of the question we are trying to answer here – is that there are differences, but in both cases transfers and in-kind incomes are a critical part of the equation.

In our entry on world poverty you can read more about the reliability of national-account poverty estimates vis-a-vis household-survey estimates.

Linking household survey (HS) and national accounts (NA) concepts of consumption – Atkinson (2016)

What we have done in the last 6 months

With a new year ahead it is a good time to look back at our work in the last months.


These were the highlights for me:


  • Thanks to Jaiden Mispy, who also joined in 2016, the web-framework of our publication got much better.

Most importantly he redesigned the OWID-grapher, the presentations and the WordPress theme. Speed is improved and he managed to make the page useful on mobile phones and tablets.
He set up proper caching (through  Cloudflare) so that the site can serve thousands of people and still be fast. And he made a series of stability improvements, bugfixes and small feature additions to interactive visualizations.
Jaiden is already working on further improvements, currently he is working on labelled timeline scatterplots.


  • We were very happy that in the new academic year many used our material for teaching. We saw that the page got accessed at many universities, and we got very nice feedback from students and lecturers.
  • I collaborated with the team from ‘Kurz Gesagt’ to make a video on population growth.
  • We launched – a short presentation on how Africa is changing.


I am very much looking forward to the work on Our World in Data in 2017! If you have requests and ideas let us know – there is a feedback option below or just send us an email.

Big thanks to Jaiden and Esteban for their work! It is great working in a team!

And big thanks to all of you who donated! It was only thanks to your support that we could continue working on Our World In Data.

Recent improvements in cross-country taxation data

This blog post relates to our entry on taxation, specifically the section discussing data quality.

Having access to reliable cross-country data on taxation is important because it helps us understand and contextualize the changing landscape of public policy around the world. And yet, while the importance of high-quality data on taxation that is comparable across time and countries is widely recognised, there are substantial deficiencies in the estimates published by traditional sources. In this blog post I want to highlight the efforts of the International Centre for Tax and Development (ICTD) to produce a new dataset that tries to address some of these deficiencies, and that has perhaps not received all the attention it deserves.

Today, the most widely used source of cross-country data on taxation is arguably the IMF Government Finance Statistics (IMF-GFS). Unfortunately, despite IMF efforts, the estimates published in this source are problematic for a number of reasons.

Perhaps the most important limitation of IMF-GFS estimates is coverage: there is extensive missing data, with large spatial and temporal gaps. Yet coverage is not the only limitation. The IMF-GFS estimates are inconsistent with estimates published in other mainstream sources such as the World Bank World Development Indicators (WB-WDI), both because of differences in the methodology used to construct the variables, and because of inconsistencies in the way countries collect and report underlying revenue data.

The following chart, from Prichard (2016), shows an example of data discrepancies in tax revenues for Ghana. The different series correspond to different sources: the blue line denotes estimates using IMF Article IV reports, the orange line denotes estimates from the IMF-GFS, the yellow line denotes data from IMF Country Reports, and the green line denotes estimates from the World Bank World Development Indicators.

As can be seen, research findings and policy conclusions would be quite different depending on the source that one uses to characterize the evolution of tax revenues in Ghana.

Alternative measures of Tax revenue (% GDP), Ghana, 1980-2010 – Prichard (2016)


Considering the above, the International Centre for Tax and Development (ICTD) recently started producing a depurated dataset that combines data from multiple sources (including the OECD, the IMF-GFS and the WB-WDI among others), applying consistency checks and flagging potential problems that may arise with the interpretation of estimates. This is a very valuable resource for research and a clear improvement over the individual underlying sources.

The following visualization maps cross-country estimates of total tax revenues as a share of GDP from the new ICTD Government Revenue Dataset (ICTD-GRD). You can find many other visualizations using this data in our entry on taxation.

One of the key advantages of the ICTD-GRD is that it stipulates a clear hierarchy for underlying sources, country by country. Generally speaking, the ICTD-GRD prioritises sources with (i) longer time series, and (ii) higher levels of disaggregation of sub-categories of revenues, such as social contributions. Yet each case is analysed independently, so the underlying sources are hand-picked, country by country, year by year. This yields fewer missing observations and more consistent estimates across countries and time.

Another important advantage of the ICTD-GRD is that it flags estimates that seem problematic. These flags include remarks such as ‘Data not credible, ‘Data is of Questionable Analytical Comparability’, and ‘Cannot Exclude Resource Revenue from Sub-Components of Total Tax Revenue’. This is evidently convenient for anyone interested in doing econometric analysis: researchers can report results both with and without the flagged data.

Of course, there remains much to be done regarding cross-country data quality in the field of taxation. While the transparent process of manual data cleaning in the ICTD-GRD is clearly a step forward, it is by no means a definitive solution. Misreporting by countries makes it difficult to unambiguously rank sources.

You can read more about how the ICTD-GRD was constructed in Prichard et al. (2014). And you can read more about how this new dataset has been used to re-examine major research questions – such as the relationships between tax and aid, elections, economic growth, and democratization – in Prichard (2016).