6. Life expectancy and length of life inequality in the long run

Lamar Crombach
ETH Zürich, KOF Swiss Economic Institute
Jeroen Smits
Radboud University, Global Data Lab
Christiaan Monden
University of Oxford

The increase in life expectancy (LE) over the last 150 years is one of humankind’s greatest achievements. Throughout the 19th century, LE at birth in the best-performing countries was less than 60 years (Maddison, 2001[1]). Nowadays, LE at birth is over 85 for women and over 80 for men (Oeppen and Vaupel, 2002[2]), and there are good reasons to expect this increase to continue in the coming decades1 (Oeppen and Vaupel, 2002[2]; Vaupel et al., 1998[3]). The intuitive appeal of the concept of LE and its wide availability make it one of the most important indicators of the overall performance of societies (UNDP, 2017[4]).

While LE gives an excellent indication of the longevity of the members of a population, it is not informative of another fundamental aspect of length of life: the degree to which the available years of life are distributed equally among the population members. Length of life inequality (LI) is one of the most fundamental forms of inequality. Whereas other forms of inequality, such as in income, wealth, education or occupation, might be compensable using redistributive policies, a high level of LI indicates that a substantial share of the population has died prematurely, a situation from which no recovery is possible (Pradhan, Sahn and Younger, 2003[5]; Smits and Monden, 2009[6]). Redistributive policies might improve the situation of the current population but not for those who died prematurely in the past.

This chapter studies historical trends in both life expectancy and life inequality of those aged 15 and over. These indicators should be studied simultaneously, as they are highly correlated. Increases in LE due to, for instance, improvements in health care will in most cases be associated with a decrease of LI. If this correlation is not accounted for, changes in LI may, largely, reflect changes in LE (Smits and Monden, 2009[6]). Studying LI while taking into account the effect of LE thus provides additional valuable information.

The next section will discuss the central concepts in more detail. Thereafter, the historical data sources are described, followed by an assessment of their quality. We then present the main trends in the indicators for the different countries and regions of the globe over the last 150 years and then assess their correlations with GDP per capita (GDPpc). The final section will discuss priorities for future research.

Life expectancy (LE) – also called period life expectancy – is “the age that a person of a particular age is expected to reach based on the age-specific mortality rates prevailing at a specific point in time” (van Zanden et al., 2014[7]). LE does not accurately reflect how long individuals may expect to live, as it reflects the mortality rates prevailing in the year when they are born, rather than the rates experienced over the course of their life. It does, however, provide an overall indication of the health performance of a society at that moment in time (Smits and Monden, 2009[6]).

We illustrate the meaning of length of life inequality (LI) by depicting the distribution of length of life for countries at different levels of development. Figure 6.1 does this for men in Niger, Brazil and Japan in the year 2000. This figure, derived from Smits and Monden (2009[6]), shows that the distribution generally has two peaks. For all countries, there is a first peak at the start of life, reflecting the relatively high mortality at birth or soon thereafter. This peak is high in Niger and low in Japan, which reflects the strong reduction of infant and child mortality that countries experience in the course of their modernisation process. From the age category of 5-10 onwards, the number of individuals reaching a given age gradually increases until a second peak occurs for the group aged 65 and over. After this peak, the number of individuals reaching an older age quickly decreases. The difference in LI among the three countries is clearly reflected in the variation of the distributions surrounding the old age peak, which is much larger in Niger than in Japan.

This chapter focuses solely on the old age peak, because the two peaks are to a certain extent a reflection of different underlying processes, e.g. Edwards and Tuljapurkar (2005[8]). A reduction of infectious diseases, mainly due to improved sanitary practices and the diffusion of new and more effective medicine, is the main cause of the reductions of infant and child mortality that countries experience. Such developments, however, have less influence on adult mortality patterns, which change at a slower pace and are partly determined by other factors (Bloom, Canning and Sevilla, 2003[9]; Cutler and Meara, 2004[10]; Marmot, 2005[11]). A comparison of inequality measures of a two-peaked distribution is also problematic, as one would be unable to determine whether changes in the inequality measures are a result of changes around the first peak or the second one. As we are mostly interested in distributions among adults, we have chosen to focus on individuals aged 15 and older.

As Figure 6.1 illustrates, most differences in the distribution of length of life between countries at different levels of development are on the left side of the old-age peak. This is largely due to the fact that the distribution of length of life is bounded at the right (Smits and Monden, 2009[6]). It is not bounded completely, as over time the proportion of persons who become very old grows steadily, but it is bounded in a practical sense, as the actual number of very old persons remains limited. There is no fundamental reason why human beings could not become 150 years old, but as far as we know, nobody has ever reached that age. Moreover, the technology to allow more individuals to reach old age is more readily available than the technology necessary for old-age individuals to reach an even older age.

Given this right-hand side boundedness of the distribution, most of the dynamics in the distribution of length of life take place left of the second peak. This implies that, with increasing LE, length of life becomes more and more concentrated in a small age band around the second peak. As such, the variation in length of life diminishes as LE increases. Indeed, previous research has shown a strong negative correlation between LE and LI (a Pearson’s r exceeding 0.9) (Shkolnikov, Andreev and Begun, 2003[12]; Wilmoth and Horiuchi, 1999[13]; Edwards and Tuljapurkar, 2005[8]; Seaman, Leyland and Popham, 2016[14]; Németh, 2017[15]; Smits and Monden, 2009[6]). This raises the question whether changes in LI as such offer much new insight into social inequality, beyond what is already known from changes in LE.

The concept of relative length of life inequality (RLI), introduced by Smits and Monden (2009[6]), addresses this issue by studying variation in LI between countries at similar levels of LE. In this manner, an indicator of LI is obtained that is independent of LE. In this chapter, this approach is illustrated by looking at variation in LI and in premature mortality within country-year combinations with similar levels of LE, and by showing trajectories of countries through the LE-LI space. In this way we aim to increase our understanding of the degree to which changes in LI are the result of changes in LE or of other factors, such as social distribution mechanisms.

LE is calculated as the mean age of death from age 15 onwards, while LI is measured by computing the Gini coefficient over the distribution of age at death from the same population. The distributions of age at death were obtained by applying the age- and sex-specific mortality rates from the life table to a population of 100 000 individuals aged 15, thus standardising for differences in adult population structure among countries and time periods. In the remainder of this chapter, LE and LI refer to the population aged 15 and over. Premature mortality is defined as the sum of all life-table deaths per 1 000 in the 15-50 age group (Smits and Monden, 2009[6]).

The data needed for the computation of LE and LI consists of period life tables (LTs). These are tables with information on the total population and the number of deaths for a specific region or country in a specific time period. Given the simplicity of the required information, this kind of data has been routinely collected by statistical offices of many countries for centuries. For this reason, LT series in many (particularly European) countries date back to the 19th, 18th or even 17th century. As there are substantial differences between the mortality patterns of men and women, separate LTs for the two groups are generally constructed.

For developed countries, LTs are regularly produced by statistical offices and national health monitoring organisations on the basis of vital and population statistics information. The most important source of life tables is the Human Mortality Database (HMD) through which a large number of high-quality life tables is made available. These life tables have been constructed according to a standardised procedure from birth and death counts derived from vital statistics and population data from censuses and official population estimates. The HMD provides life tables for about 40 countries, including long time series that sometimes date back to the 19th or even 18th century (HMD[16]). The HMD is the main data source used in this chapter. For countries not represented in the HMD, a second preferred source is life tables derived directly from statistical offices. These life tables are generally also of good quality, but because of differences in methods and data quality they might be less comparable than those derived from the HMD.

Because low- and middle-income countries are not well represented in the HMD, and because the statistical offices of these countries generally do not publish life tables, we had to rely on other sources for them. In order of preference, we used life tables from the Global Health Observatory of the World Health Organization (WHO, 2018[17]), from the Human Lifetable Database (Shkolnikov, 2017[18]), or from the GDL Length of Life Database (Smits and Monden, 2009[6]). In total, our database includes 15 144 sex-specific LTs, of which 7 582 for men and 7 562 for women.

The time period for which the deaths are counted in an LT is often a year, but it may also be a shorter or longer period. For the population size of the age groups, the mid-year population is generally (but not always) used. The size of the age groups is usually one or five years. When five-year groups are used, the first two groups deviate from the general pattern, as the first group includes only children under the age of one and the second group only children aged one to four. With the third group aged five to nine, the regular five-year pattern starts. The highest group is always an open-ended group that includes all persons of and above the highest age (e.g. 85+, 95+, 100+, 110+).

To achieve comparability between the different sources, all LTs used in this chapter are abridged (85+) period LTs, using 5-year age intervals. We chose to work with abridged 85+ LTs, because for part of the countries and time periods it is the only type available. Additionally, we wanted to include as many LTs as possible to achieve the widest possible coverage. LTs with more detail in our database were recalculated into the abridged 85+ form to make them comparable to the others. Hence LI among the eldest (85+) age group is not reflected in our results.

In total, our database includes 15 144 life tables for 203 countries from all regions of the world. The database is available at the GDL Length of Life website (Global Data Lab[19]).

An overview of the quality of the data sources is provided in Table 6.1. As with data from other historical sources, life tables are not without problems. Although the required data (number of deaths and total population) are not very complex, the quality of the available records depends on the state of development and procedures used by administrative offices, which might have differed considerably between countries and regions. Besides such obvious differences in the original data sources, the quality of the life tables that are used in this chapter also varies depending on the method that was used to construct them.

As discussed above, the 6 886 life tables derived from the Human Mortality Database are of highest quality (Rank 1 in Table 6.1, as they are constructed according to a standardised method from the source material. Life tables from the HMD are therefore fairly comparable between countries and years.

The life tables of national statistical offices and most other sources are more varied and therefore of somewhat lower quality (Rank 2). This is also true for the life tables from the Global Health Observatory of the World Health Organization, which includes the best possible values and estimates for all WHO member states except for a few island states (WHO, 2018[17]). The quality may vary however between countries and regions, and sometimes life tables are estimated. This also applies to the Human Lifetable Database (Max Planck/Berkeley), which contains published life tables from a broad range of sources for which no further information on data quality is available (Shkolnikov, 2017[18]). Both sources are therefore indicated with Quality Rank 2 in Table 6.1. Of the life tables used for this chapter, 7 984 are of Rank 2. The remaining 324 life tables from other – miscellaneous – sources derived from the Smits and Monden (2009[6]) life table database are indicated with Quality Rank 3.

Figure 6.2 provides an overview of LE (15+) for men (left) and women (right) for the period 1800-2016. We see that, over time, the number of LTs has increased gradually until 2000. A large increase in the number of LTs is observable since 2000, when our dataset starts to include life tables from the WHO for all the world’s countries. The strong increase of observations (country-years) with a low level of life expectancy is due to the fact that the WHO database contains many life tables for low- and middle-income countries, for which none or only a few LTs were available earlier.

Until 1900, LE in the best-performing countries was less than 65 years for men and less than 67 years for women. For men, life expectancy then increased until about 72 around 1950, plateaued at this level until about 1970, and then increased gradually until its current value of a little bit above 82. For women, a similar pattern can be observed.

Periods of war are observable in the left panel, showing severe drops in LE for men in selected years. Clear examples are the Napoleonic wars in the 1803-1815 period, including his 1812 campaign into Russia, and the First and Second World Wars. For women, wars have less influence on LE than for men, but the disastrous Spanish flu of 1918 is clearly visible.

Table 6.2 and Table 6.3 provide an overview per decade of LE for all men and women in the 25 countries on which this book focuses for the period 1800-2016. One can observe a large degree of variation between countries and over time. The tables show that LE has increased in all 25 countries. A man living in Sweden in the first decade of the 19th century was expected to live for about 57 years, but in the 2010s he is expected to live 80.4 years. Other countries have made similar gains, though at a different pace. It took Sweden about two centuries to improve LE by 23 years. This same improvement was achieved in India in about half the time – a clear indication that less developed countries are leapfrogging the stages of health innovation as late adopters of new technologies. However, many countries have not yet reached the final stage, as large differences in LE across countries persist in the 2010s. Although India has improved at a fast pace, its LE for men is currently still 70.3 years, compared to the LE of 80.9 of the current frontrunner Australia. This is a difference of 10.6 years. African countries lag even further behind, with an LE of 61.7 for men and 62.4 for women in Nigeria, and an LE of 62.0 for men and of 68.1 for women in South Africa. As such, there is still great potential for improvement.

Some variation also exists amongst high-income countries. While Australia and Canada have the highest male LE in the 2010s (of 80.9 and 80.6 years respectively), male LE in the United States and Germany is clearly lower (77.2 and 78.5 respectively). For women, all Western European and Western Offshoot countries have an LE of over 83 in the 2010s, while the United States’ LE remains below 82.

Regarding gender differences, the figures reveal higher LE for women than for men in almost all cases. The most extreme case is Russia, where women on average lived 11.1 years longer than men in the 2010s, and 12.8 years longer in the 2000s: a staggering difference. In Nigeria, women also lived longer than men in the 2010s, but this difference was limited to only 0.7 years. In some exceptional cases, men lived longer than women, as in India in the period between the 1900s and the 1970s, reflecting the “missing women” phenomenon noted by Sen (1990[20]).

Periods of war are clearly observable as well. A large drop in LE of 10.8 years is observed in the 1910s for French and Italian men during World War I. We also see a drop in LE in the 1940s during World War II in the Netherlands, France and Italy. After World War II, LE recovered rather quickly and since then shows a steady increase over time, with a faster pace in less developed countries.

Figure 6.3 shows the Gini coefficients for length of life inequality (LI) in the population aged 15 and over for the period 1800 to 2016. The high correlation between LE and LI (discussed in Section 6.2) is clearly visible, as Figure 6.3 (on length of life inequality) almost exactly mirrors Figure 6.2 (on average life expectancy).

The Gini coefficient for length of life inequality for men runs from about 0.07 to 0.30 and for women from 0.04 to 0.25. These are low levels compared to Gini values for income inequality at the country level, which run from about 0.25-0.30 in the best-performing countries to about 0.60-0.65 in the worst-performing countries (Solt, 2019[21]). This means that the distribution of years of life in the 15+ population is clearly narrower than the income distribution. This does not mean, however, that length of life inequality is an unimportant issue. While income inequality can be compensated using redistributive policies, a high value of LI means that people have died at too young an age, a situation that cannot be compensated.

Figure 6.3 indicates that LI has decreased over time. This is confirmed by Table 6.4 and Table 6.5, which show the trends in LI for the 25 countries in the period 1800-2016. In the large majority of cases, we see a clear decrease of LI. There are some notable exceptions in the 2000s and 2010s, when Mexico and Egypt faced increased inequality. Over the past few decades, changes in LI have been rather small in India, Indonesia, Thailand, Argentina, Brazil and Mexico. In the developed world, the United States has a relatively large LI compared to other developed countries, an issue that is discussed in more detail later in this chapter.

It remains to be seen what the decrease of length of life inequality observed for most countries exactly means. As discussed earlier in this chapter, there exists a strong negative correlation between the trends in LE and LI, with LI almost automatically decreasing when LE increases. This correlation, which is -0.94 for men and -0.97 for women in our data, is clearly visible in Figure 6.4, which depicts the relationship between LI and LE without the time dimension. Each dot in this figure represents the values of LE and LI for a specific country in a given year; the left figure is based on 7 582 life tables for men and the right figure on 7 562 life tables for women.

The extremely strong association between LE and LI, as depicted in Figure 6.4, raises the question whether changes in LI as such may offer many new insights beyond what is known from changes in LE. Smits and Monden (2009[6]) answer this question confirmatively, by pointing to the fact that at each level of LE there is substantial variation between observations (country-years) with more and with less inequality. In their data, the number of premature deaths in the 15-50 age group is significantly and considerably higher in the 20% most unequal country-years compared to the 20% least unequal country-years for a given level of LE. They therefore introduce the concept of Relative Length of Life Inequality (RLI), which refers to the variation in LI between observations with a similar LE.

To gain insight into the importance of this pattern in our data, Figure 6.5 zooms in on it by providing information on length of life inequality and on premature deaths within LI quintiles, for two levels of average life expectancy: an LE of 65 (i.e. 64.5-65.5) and an LE of 75 (i.e. 74.5-75.5). The left panel shows that there is substantial variation in LI across observations (country-years). For a life expectancy of 65 (the top lines in the panel), length of life inequality varies between 0.139 and 0.179 for women (shown in red) and between 0.125 and 0.165 for men (shown in blue). For a life expectancy of 75, it varies between 0.079 and 0.111 for women, and between 0.084 and 0.110 for men. All these differences are significant at p<0.001, which makes clear that they are not negligible.

To gain insight into the importance of this pattern in our data, Figure 6.5 zooms in on it by providing information on the variation in length of life inequality and premature deaths between countries with higher inequality and countries with lower inequality at the same level of LE. The left panel (A) shows how LI varies between the lowest, 2nd, 3rd, 4th and highest LI quintiles at an LE of 65 (64.5-65.5) and at an LE of 75 (74.5-75.5). For a life expectancy of 65 (the top lines in the panel), length of life inequality varies between 0.139 and 0.179 for women (shown in red) and between 0.125 and 0.165 for men (shown in blue). For a life expectancy of 75, it varies between 0.079 and 0.111 for women, and between 0.084 and 0.110 for men. All these differences are significant at p<0.001, which makes clear that they are not negligible.

A similar pattern is visible in the right panel of Figure 6.5, which shows premature mortality – as measured by the number of deaths per 1 000 people between age 15 and 50 – for the two life expectancy levels. At both levels of LE, premature mortality in this age group varies considerably between the lowest, 2nd, 3rd, 4th and highest LI quintiles. At an LE of 65, it varies between 168 and 297 deaths per 1 000 for women, and between 132 and 247 deaths per 1 000 for men. At a life expectancy of 75, it varies between 37 and 95 deaths per thousand for women, and between 43 and 94 deaths per thousand for men. Hence, in both cases, the most unequal quintile has substantially more premature deaths than the most equal quintile, both for women and for men. All differences are again significant at p<0.001.

An effective way to increase our understanding of the dynamics of inequalities in the age of death is to highlight the trajectories described by individual countries through the dotted areas in Figure 6.4, hence against the background of the LE-LI values for all other country-year combinations in our database. In Figure 6.6, this is done for four countries – the United States, France, Sweden and Japan. Each dot represents a gender-specific LE-LI value of a specific country in a specific year. Values for men are in blue, values for women in red.

The country-specific trajectories through the LE-LI space are not steadily decreasing chronological lines. During wars or epidemics, when LE may drop and LI increase, a country tends to jump in the upper-left direction. In the following years, when things go back to normal, the country may jump back to around its earlier position, to continue its path towards (on average) a higher LE and lower LI.

Figure 6.6 shows that during much of its history, the United States has been at an average level of LI given its LE, but that at the higher levels of LE (above an LE of 74) the country tended to move upward in the LE-LI space. France had a high level of length of life inequality for the lower values of average life expectancy, but it had lower inequality between LE 55 and LE 70. After LE attained levels of 75 for males and 80 for females, the decrease of LI decelerated, and the country has moved upwards in the LE-LI distribution.

Japan’s trajectory is less clear-cut, as the trajectories of men and of women differ. Until an LE of about 74, inequality among Japanese men was relatively low, while among Japanese women it was much higher. After LE 74, the pattern reversed, with inequality among men being higher than among women. However, while for Japanese men the highest LE is associated with an average level of LI, Japanese women were at the upper end of the distribution when reaching their highest level of LE.

In the case of Sweden, for an LE between 50 and 60, both men and women enjoyed one of the lowest length of life inequalities. However, when LE grew beyond 60, the decrease in LI slowed down until LE 68, when Sweden was among the most unequal countries within this LE group. From LE 68 on, further increases in LE were associated with substantial decreases in LI, particularly for men. At the highest level of LE, Swedish men are therefore among the most equal populations in their LE group, whereas Swedish women are in the middle.

The comparatively high level of LI among women in times of low LE is most likely due to high maternal mortality rates associated with pregnancy and childbirth, which until the 1960s contributed substantially to premature mortality among women, even in the most developed countries. The strong reduction of this form of mortality most likely is responsible for the crossing of the trajectories of women and men somewhere around an LE of 70.

In this section, trends in life expectancy and length of life inequality are analysed alongside trends in GDP per capita. Correlation coefficients for each decade are shown in Figure 6.7 for life expectancy, and Figure 6.8 for length of life inequality, respectively.

Figure 6.7 shows that the correlation between average life expectancy and GDP per capita was weak during most of the 19th century, if it existed at all. A clear positive correlation emerges only around 1820-1830. The wide confidence intervals reflect the differences in data availability over time. After 1900, the positive correlation becomes stronger, on the order of 0.3-0.5, which means that countries with higher GDP per capita had higher levels of LE. This could indicate that higher life expectancy leads to more productivity, or that an increase in income allows individuals to live healthier – and consequently longer – lives.

An additional explanation for the fact that the correlation between LE and GDP per capita was relatively strong in the 1910s-1950s period is World War I and World War II, as large drops in LE were accompanied by large drops in GDP per capita. The correlation then weakened up until the 1970s, reflecting a return to normal. The correlation then starts to strengthen again, up to the same high levels as during World War II. A possible explanation for this is the rise of the welfare state, allowing for increases in GDP per capita to lead to stronger increases in LE, as populations started to gain access to health care.

Figure 6.8 presents the correlation between length of life inequality and GDP per capita. As in the case of LE, we observe a shift in the nature of the correlation over time. During the 19th century, the correlation was mostly significant and positive, indicating that higher GDP was associated with higher levels of LI. This might be explained by the fact that, in the earlier phase of industrialisation, the GDP per capita of industrialising countries was growing substantially, but this came at the cost of the poorer health of the mass of workers, reflecting low labour standards and living in large cities with poor quality housing, bad sanitation and weak public health facilities (Fogel, 2004[22]; Steckel, 2001[23]).

From the 1890s onwards, the correlation turns negative, with more developed countries showing lower values of LI. This might reflect an improvement in labour conditions and public (health) services in these countries. The negative association peaked in the 1930s at a relatively strong level of -0.6, after which it decreased again to around -0.3 in the 1970s. Since then it has increased again to a value of about -0.6 in the 2010s. In other words, from the 1930s onwards, the association between GDP per capita and LI weakened, up until the 1970s. From the 1970s onwards, the association started to regain strength. This is in line with the observed pattern of the correlation between LE and GDP per capita. During the First and Second World Wars, LI increased strongly as a result of the loss of soldiers, while GDP per capita decreased strongly. The introduction of the welfare state in the 20th century may have translated part of the average increases in GDP per capita into better health for the more vulnerable part of the population, leading to a lower LI.

Life expectancy and length of life inequality are among the most fundamental indicators of the status of human societies. A low LE compared to what is potentially possible means that more people die than is necessary. Moreover, a high level of LI within a country means that too many people die at too young an age. Given that death is irreversible, and that the deceased cannot be compensated for their disadvantage, LE and LI are directly linked to the success or failure of societies in satisfying their citizens’ basic needs. As such, these indicators should be among the first to consider when assessing the performance of societies.

Given the extremely high correlation between LE and LI, it is essential that, besides the absolute value of LI, its relative value, for a given value of life expectancy, is also considered. As LI more or less automatically decreases as countries achieve higher LE, the major question is not what the level of LI in a country is, but to what extent the country’s level of LI is higher or lower than that of other countries at the same level of LE. As shown in Figure 6.5, the values of LI and the number of premature deaths can vary greatly between countries at similar levels of LE. A high level of LI, or premature deaths, at a similar level of LE suggests that not enough resources are being spent on improving the health situation of the younger and middle-aged population groups. As such, these inequality measures can provide a broad indication of distribution problems related to the population’s health within a society.

The value of LE and LI as indicators of societies’ performance is strengthened by their ease of construction. All one requires is information on the male and female population and deaths within age categories in a specific time-period. Hence, the required information is very factual and more easily observable than that for most other indicators, such as, for instance, (national) income or morbidity.

Given the fundamental character of indicators related to life and death, and the simplicity of their measurement, the registration of population data already started centuries ago in more industrialised societies. Official population registrations date back to the 19th century in many European countries, and even to the 1750s in Sweden and Finland. Given that over the whole period since then the same data has been collected, these registrations offer us the possibility to look further back in time than is possible with data for most other indicators.

This does not mean, however, that the population-based measures presented in this chapter are without problems. Besides the obvious measurement issues associated with historical data, there is a more fundamental problem related to the use of these measures to indicate societies’ performance: They suffer from a time lag, as their current values are determined by the situation and behaviour of people in the past. Food scarcity during childhood or smoking behaviour during young adulthood influences old-age mortality that is used for computing LE for later-born cohorts.

This problem to a certain extent also exists for LI, as determining inequality in length of life involves a comparison between mortality in older age groups with mortality at a younger age. However, given that a relatively high level of LI at a certain level of LE indicates that more people die at too young an age, this indicates that not enough resources are being spent on improving the health situation of the younger and middle-aged population groups. Hence, by combining LI with LE, a broad indication of distribution problems related to population health within a society can be obtained.

We believe that even more is possible, and that substantial progress can be achieved without great additional data collection efforts. The life tables used for constructing LE and LI contain more information that can be used for this purpose. They provide, for each five-year age category, and for men and women separately, the group-specific mortality rates. These mortality rates can be compared between different age groups within the same country, as well as between the same age groups in different countries and years. In this way, we are able to observe – very specifically and currently – which age groups in which countries and years do worse – or better – than would be expected based on what is observed in other situations.

As a priority for future research, we therefore recommend developing a dashboard based on these age-specific mortality rates, with which the situation and life chances of men and women in the different age groups can be monitored. The life table database underpinning this chapter offers ample possibilities for conducting cross-national and cross-temporal comparative research to validate such a dashboard by determining which societal situations are associated with deviant patterns in the different phases of life.

References

[9] Bloom, D., D. Canning and J. Sevilla (2003), The Demographic Dividend: A New Perspective on the Economic Consequences of Population Change, RAND, http://dx.doi.org/10.7249/MR1274.

[8] Edwards, R. and S. Tuljapurkar (2005), “Inequality in life spans and a new perspective on mortality convergence across industrialized countries”, Population and Development Review, Vol. 31/4, pp. 645-674, http://dx.doi.org/10.1111/j.1728-4457.2005.00092.x.

[22] Fogel, R. (2004), The Escape from Hunger and Premature Death, 1700-2100: Europe, America and the Third World, Cambridge University Press, http://dx.doi.org/10.1017/CBO9780511817649.

[19] Global Data Lab (n.d.), Length of Life Database, https://globaldatalab.org/lengthoflife/ (accessed on 14 January 2020).

[16] HMD (n.d.), The Human Mortality Database, https://www.mortality.org/ (accessed on 14 January 2020).

[24] Kontis, V. et al. (2017), “Future life expectancy in 35 industrialised countries: Projections with a Bayesian model ensemble”, The Lancet, Vol. 389/10076, pp. 1323-1335, http://dx.doi.org/10.1016/S0140-6736(16)32381-9.

[1] Maddison, A. (2001), The World Economy: A Millennial Perspective, Development Centre Studies, OECD Publishing, Paris, https://dx.doi.org/10.1787/9789264189980-en.

[11] Marmot, M. (2005), “Social determinants of health inequalities”, The Lancet, Vol. 365/9464, pp. 1099-1104, http://dx.doi.org/10.1016/S0140-6736(05)71146-6.

[25] Mathers, C. et al. (2015), “Causes of international increases in older age life expectancy”, The Lancet, Vol. 385/9967, pp. 540-548, http://dx.doi.org/10.1016/S0140-6736(14)60569-9.

[15] Németh, L. (2017), “Life expectancy versus lifespan inequality: A smudge or a clear relationship?”, PLoS One, Vol. 12/9, http://dx.doi.org/10.1371/journal.pone.0185702.

[2] Oeppen, J. and J. Vaupel (2002), “Broken limits to life expectancy”, Science, Vol. 296/5570, pp. 1029-1031, http://dx.doi.org/10.1126/science.1069675.

[5] Pradhan, M., D. Sahn and S. Younger (2003), “Decomposing world health inequality”, The Journal of Health Economics, Vol. 22/2, pp. 271-293, http://dx.doi.org/10.1016/.

[14] Seaman, R., A. Leyland and F. Popham (2016), “Increasing inequality in age of death at shared levels of life expectancy: A comparative study of Scotland and England and Wales”, SSM - Population Health, Vol. 2, pp. 724-731, http://dx.doi.org/10.1016/j.ssmph.2016.10.001.

[20] Sen, A. (1990), “More than 100 million women are missing”, New York Review of Books, pp. 61-66, https://www.nybooks.com/articles/1990/12/20/more-than-100-million-women-are-missing/.

[18] Shkolnikov, V. (2017), “Methods protocol for the human life-table database”, Working Paper, https://www.mortality.org/Public/Docs/MethodsProtocol.pdf.

[12] Shkolnikov, V., E. Andreev and A. Begun (2003), “Gini coefficient as a life table function: Computation from discrete data, decomposition of differences and empirical examples”, Demographic Research, Vol. 8/11, pp. 305-358, http://dx.doi.org/10.4054/DemRes.2003.8.11.

[6] Smits, J. and C. Monden (2009), “Length of life inequality around the globe”, Social Science & Medicine, Vol. 68/6, pp. 1114-1123, http://dx.doi.org/10.1016/j.socscimed.2008.12.034.

[21] Solt, F. (2019), “Measuring income inequality across countries and over time: The standardized world income inequality database”, Social Science Quarterly, Vol. 101/3, pp. 1-23, https://doi.org/10.1111/ssqu.12795.

[23] Steckel, R. (2001), Industrialization and health in historical perspective, Oxford University Press, http://dx.doi.org/10.1093/acprof:oso/9780192631961.001.0001.

[26] UN (2019), World Population Prospects 2019: Data Booklet, Statistical Papers - United Nations (Ser. A), Population and Vital Statistics Report, United Nations, New York, https://dx.doi.org/10.18356/3e9d869f-en.

[4] UNDP (2017), “Human Development Report 2016”, United Nations Development Programme, http://dx.doi.org/10.18356/6d252f18-en.

[7] van Zanden, J. et al. (eds.) (2014), How Was Life?: Global Well-being since 1820, OECD Publishing, Paris, https://dx.doi.org/10.1787/9789264214262-en.

[3] Vaupel, J. et al. (1998), “Biodemographic trajectories of longevity”, Science, Vol. 280/5365, pp. 855-860, http://dx.doi.org/10.1126/science.280.5365.855.

[17] WHO (2018), “WHO methods and data sources for life tables 1990-2016”, World Health Organization, https://www.who.int/healthinfo/statistics/LT_method.pdf.

[13] Wilmoth, J. and S. Horiuchi (1999), “Rectangularization revisited: Variability of age at death within human populations”, Demography, Vol. 36/4, pp. 475-495, http://dx.doi.org/10.2307/2648085.

[10] Wise, D. (ed.) (2004), “Changes in the age distribution of mortality over the twentieth century”, University of Chicago Press.

Length of life inequality (LI) is measured by computing the Gini coefficient over the distribution of age at death based on the population aged 15 and over. The distributions of age at death were obtained by applying the age- and sex-specific mortality rates from the life table to a population of 100 000 individuals aged 15, thus standardising for differences in adult population structure among countries and time periods. The Gini coefficient varies between zero and one, with zero indicating a situation of maximum equality (everybody has the same age at death) and one indicating maximum inequality. The choice of the inequality measure for computing length of life inequality is not a very critical one. There is a high correlation (generally over 0.90) among inequality measures computed over the distribution of age at death, and the use of different inequality measures leads to similar results (Wilmoth and Horiuchi, 1999[13]).

Note

← 1. Recent sources confirm a projected increase in the limits of LE, albeit at a potentially slower pace (Kontis et al., 2017[24]; UN, 2019[26]). The main causes for these increases are reduced tobacco usage and a reduction in cardiovascular disease mortality (Mathers et al., 2015[25]). There is, however, large variation among developed countries in terms of LE trends. For instance, the United States, which has a relatively low level of LE when compared to other developed countries, is projected to face declines in LE (Kontis et al., 2017[24]).

Metadata, Legal and Rights

This document, as well as any data and map included herein, are without prejudice to the status of or sovereignty over any territory, to the delimitation of international frontiers and boundaries and to the name of any territory, city or area. Extracts from publications may be subject to additional disclaimers, which are set out in the complete version of the publication, available at the link provided.

© OECD 2021

The use of this work, whether digital or print, is governed by the Terms and Conditions to be found at http://www.oecd.org/termsandconditions.