We use data from nationally representative datasets that assess the health, social and economic conditions of non-institutionalised older adults, with the majority being sister studies of the Health and Retirement Study (HRS) (National Institute on Aging, 2007). Moreover, the main contents of the surveys are harmonised to enable cross-country comparisons (Smith, 2021). All the considered surveys include comparable performance tests on cognitive or physical functioning (Phillips et al., 2023; Shih et al., 2012), which are recognised indicators for older adults’ ability to live an independent life (Matsui et al., 2014; Salthouse, 2012). In this study, we consider the most recent waves of the following 11 surveys (presented alphabetically), which represent 41 countries across the world.

Table 1 provides a compact overview of all the surveys, countries, wave years and functioning measures (i.e., HGS, immediate recall and verbal fluency) included in the study. We include in our sample all respondents aged between 50 and 84 who performed at least one of the three functioning tests and provided information on their age and gender, stating that they were either female or male. For data availability reasons, we have chosen 84 years as the upper bound.

Before detailing the variables used for cognitive and physical functioning, we provide in the following an overview of each of the 11 surveys, particularly focusing on their data collection design.

The Chile Cognitive Aging Study (Chile-Cog) commenced its longitudinal study in 2019. Its main objective is to measure the prevalence of dementia and cognitive ability among the Chilean population aged 60 and older. Chile-Cog follows the Harmonized Cognitive Assessment Protocol (HCAP), which was first utilised in the HRS. We use data from the 2019 Chile-Cog (Ministry of Labor and Social Welfare, 2019).

The English Longitudinal Study of Ageing (ELSA) is designed to represent the population of non-institutionalised individuals aged 50 and older living in England. The first wave was conducted in 2002, and follow-up waves were conducted biennially (Steptoe et al., 2013). Since the latest handgrip strength (HGS) assessments were performed across waves 8 and 9 (conducted in 2016 and 2018, respectively), our results are based on both waves.

The Brazilian Longitudinal Study of Aging (ELSI-Brazil), based on the HRS, is a longitudinal study of community-dwelling adults aged 50 or older living across the five great geographic regions of Brazil (Lima-Costa et al., 2022). ELSI-Brazil started in 2015 and commenced with a second release in 2019, which we use in our study.

The HRS, which was first conducted in 1992 in the United States, spans 16 finished waves (from 1992 to 2022), and is one of the earliest and most renowned longitudinal studies focusing on health, retirement and ageing (National Institute on Aging, 2007; Sonnega et al., 2014). Numerous ageing studies across the world build upon HRS and use comparable questionnaires as well as cognitive and physical functioning tests. We use data from the 15th wave, collected in 2020 (Health and Retirement Study, 2023).

The Indonesia Family Life Survey (IFLS) started in 1993 and focuses on individuals aged 50 and older, including their spouses, living in Indonesia. We use data from the most recent IFLS wave 5, which was conducted between 2014 and 2015 (Strauss et al., 2016).

The Japanese Study of Aging and Retirement (JSTAR) was conducted by the Research Institute of Economy, Trade and Industry (RIETI), Hitotsubashi University, and the University of Tokyo. JSTAR encompasses longitudinal data on middle-aged and elderly Japanese individuals aged 50 and older from 2007 onwards, with the most recently released wave 3 collected in 2013 and 2014. In this article, the results are based on JSTAR wave 3.

The Longitudinal Ageing Study in India (LASI) follows older adults aged 45 and older in India (Perianayagam et al., 2022). We use the first wave, for which data were collected between 2017 and 2018, as well as between 2020 and 2021 (Chien et al., 2023).

The Mexican Health and Aging Study (MHAS) is a national longitudinal study of adults aged 50 and older in Mexico (Wong et al., 2017). The baseline survey was conducted in 2001, with follow-up interviews performed in 2003, 2012, 2015, 2018 and 2021. In this study, we use MHAS wave 3 (2012), the latest release including the HGS test, and wave 5 (2015), the latest release including the verbal fluency test (Michaels-Obregon et al., 2023).

The Study on Global AGEing and Adult Health (SAGE) collects longitudinal data on adults aged 18 and older, with an emphasis on those aged 50 and older, from six countries: China, Ghana, India, Mexico, the Russian Federation and South Africa (World Health Organization, 2006). We use the first wave, which started in 2007 and was released for all SAGE countries (Chatterji and Kowal, 2013).

The Survey of Health, Ageing and Retirement in Europe (SHARE) has collected longitudinal data every two years on individuals aged 50 and older since its launch in 2004. With its expanding coverage across Europe and Israel, SHARE now includes 28 countries (Börsch-Supan et al., 2013). In this analysis, we use data from SHARE wave 9.

The Irish Longitudinal Study on Ageing (TILDA) is a longitudinal study that collected its initial wave in 2009. It focuses non-institutionalised individuals aged 50 and older living in the Republic of Ireland (Donoghue et al., 2018). We use data from TILDA wave 1.

The cognitive functioning tests conducted within the ageing studies included in this study cover the two dimensions of episodic memory and verbal fluency (Shih et al., 2012). Episodic memory is measured by an immediate recall test in which the interviewer reads out a list of 10 words and asks the respondent to immediately recall as many words as possible within one minute without rereading the words. In the verbal fluency test, the interviewer asks the respondent to name as many animals as possible within 60 seconds. For comparability reasons, we top-coded all scores of respondents who named more than 20 animals to 20, given that this is the lowest upper limit of the utilised studies (in ELSI-Brazil).

As a measure of physical functioning, we use the performance in a HGS test. HGS in kilograms is measured using a dynamometer, either once with the dominant hand, or more often with the weaker hand as well, mainly in a standing position (Phillips et al., 2023). For comparability reasons, we use the maximum score of the individual HGS measurements.

Figures S.1–S.3 show the distributions of the HGS, immediate recall and verbal fluency test scores for each country.

In this study, we borrow an established approach from the discipline of economics to quantify health inequality. The Gini coefficient is an indicator of statistical dispersion intended to represent the inequality within a population or subpopulation (Wagstaff et al., 1991). It ranges from zero to one, with a higher Gini coefficient indicating higher inequality. We aim to measure inequality in health by the degree of inequality in the distribution of HGS, episodic memory and verbal fluency across subpopulations. To obtain inequality estimates for different subpopulations of the investigated countries, we estimate the Gini coefficient separately for the physical and cognitive functioning variables, also considering survey weights. We obtain the Gini coefficient using the estimator G i n i ^ = 2 ∑ i = 1 n w i r i y i − ∑ i = 1 n w i y i ( ∑ i = 1 n w i ) ∑ i = 1 n ( w i y i ) − 1 (1) where y = { y i , i ∈ U } is the function distribution over the subpopulation U and w i is the survey weight, and r i the rank of i in the distribution y . We use the R package convey to obtain the Gini estimates (Djalma et al., 2023; Osier, 2009).

For each country and health measure, we estimate Gini coefficients for the subpopulation aged 50 to 84, stratified by gender and five-year age group, to gain insights into the disparities of inequalities within countries’ subpopulations. To mitigate the small sample bias of the Gini estimate, we estimate subpopulation Gini indices only for subsamples with at least 50 observations (Deltas, 2003). This exclusion criterion mostly affects the youngest five-year age group (50–54) of European countries and Israel (excluded subpopulations are indicated in Tables S.1–S.3).

To avoid comparability issues due to differences in survey design related to eligible age groups and small sample sizes, we additionally estimate Gini coefficients for the 60- to 74-year-old subsample, stratified by country, gender and five-year age group, to allow for better comparisons across countries.

Due to the low number of observations at the tails of the age distributions, we conduct the stratification by gender only for the 50–84, 60–74, 60–64, 65–69 and 70–74 age groups.

We conduct two robustness checks. First, we address the inability of the Gini index to represent specific parts of the health distribution by comparing the Gini estimates to the P80/P20 ratio. The P80/P20 ratio, also known as the quintile share ratio, is a standard measure of inequality representing the ratio of the 80th percentile to the 20th percentile of the health distribution (European Commission, 2003; Eurostat, 2005). Compared to the Gini index, the P80/P20 ratio is less susceptible to changes in the middle of common distributions, and is less impacted by their outliers. As a second robustness check, we estimate the verbal fluency Gini indices using non-top-coded verbal fluency scores and compare them to the original estimates.

For the population aged 50 to 84, the HGS Gini estimates indicate that inequality is lowest in Japan (0.172), followed by Finland (0.179) and Switzerland (0.180) (Table S.1), whereas inequality is highest in South Africa (0.281), Ghana (0.236) and Israel (0.234). Thus, in the HGS Gini indices, the difference between the minimum and maximum values is 0.109 Gini points (Table S.1).

For the cognitive functioning Gini estimates, we observe similar disparities between the minimum and maximum values. For the immediate recall test, the Gini estimates range from low values in England (0.148), Austria and Czechia (both at 0.149) to high values in Indonesia (0.268), Japan (0.236) and Brazil (0.235), yielding an immediate recall Gini range of 0.120 Gini points (Table S.2). Interestingly, while Japan ranks among the most equal countries for HGS, it is also among the least equal countries for immediate recall. For verbal fluency, Austria and Denmark (both at 0.045) are the countries with the most equal estimated distribution, followed by Czechia (0.048) (Table S.3). The highest verbal fluency Gini estimates are found in Russia (0.325), Brazil (0.262) and China (0.214) (Table S.3). The verbal fluency Gini estimates have the largest range, amounting to 0.280 Gini points.

When we limit our estimates to the population aged 60 and 74, the country rankings of the Gini indices are largely retained (Tables S.1–S.3; Figure 1).

Our results also hint at substantial differences in functioning disparities across the observed countries in different continents (Figure 1). For instance, Northern European countries and the USA exhibit lower inequalities in HGS, whereas the observed African countries and China show relatively high inequalities (Table S.1; Figure 1).

Continental differences in cognitive functioning inequalities appear to be more pronounced. The observed countries in Latin America and Southeast Asia show the highest inequalities in immediate recall, pointing to continental differences in cognitive ageing processes. Similarly, verbal fluency disparities are the most pronounced in the observed Latin American and Asian countries (Tables S.2; S.3; Figure 1).

Interestingly, the countries with lower cognitive and physical functioning Gini estimates show higher weighted median functioning scores (Figure 2). For the two cognitive functioning domains, we find statistically significant positive correlations between estimated equalities (1-Gini) and the weighted medians (p < 0.001). For HGS, the positive correlation is not statistically significant (see Figure 2).

For most countries, we estimate that women experience greater inequalities in HGS, whereas men exhibit greater inequalities in both immediate recall and verbal fluency. These observations hold true for the population aged 50 to 84 as well as for those around retirement age (60–74 years old) (Tables S.1–S.3; Figure 3).

Gender differences in the HGS Gini estimates are substantial in Bulgaria, China, Ghana and Russia, where we estimate higher inequality for women than for men, and in Israel, where inequality is higher for men than for women.

For the immediate recall Gini estimates, most countries show minimal gender differences, with the notable exceptions of Japan, where men record a higher estimated immediate recall Gini than women, and Israel, where the opposite pattern is observed. Moreover, a substantially higher verbal fluency Gini estimate for women than for men is found in Israel.

The analysis of the Gini estimates by age group suggests that there is a U-shaped relationship between the HGS inequalities and age. Specifically, the median HGS Gini across countries for the 50–54 age group is higher than that for the subsequent age groups of 55–59, 60–64 and 65–69. Beyond age 54, we observe a positive age group gradient. From the youngest (50–54) to the oldest (80–84) age group, the percentual increase of the median HGS Gini is 9.59% (from 0.188 to 0.206) (Figure S.4).

The immediate recall and verbal fluency Gini estimates show a positive age group gradient. This gradient is most pronounced for verbal fluency, with a substantial increase in the median Gini from the 50–54 to the 80–84 age group of 111.48% (from 0.097 to 0.206), compared to a 55.14% increase for episodic memory (from 0.150 to 0.233) (Figure S.4). Thus, the estimated median health inequality within five-year age groups is higher for the most advanced age group than for the younger age groups across all countries, highlighting pronounced heterogeneities among the oldest-old (Figure S.4 in the Supplementary material).

Table 2 compares age group differences in the Gini estimates for the two genders separately, focusing on the 60–64, 65–69 and 70–74 age groups. For women, we observe a consistent positive age-related gradient in the country median Gini estimates for both physical and cognitive functioning. For men, the 65–69 age group records the lowest estimates, followed by the 60–64 age group, with the 70–74 age group exhibiting the highest values across all outcomes.

Within the age groups, the median HGS Gini estimates are consistently higher for women than for men. In contrast, the median immediate recall Gini estimates are higher for men than for women. For verbal fluency, women exhibit lower median Gini estimates in the 60–64 age group but record higher values in the older 65–69 and 70–74 age groups compared to their male counterparts.

Both robustness analyses – substituting the Gini coefficient with the P80/P20 ratio and removing top-coding from the verbal fluency test score – point to the stability of our results.

When comparing the Gini with the P80/P20 ratio at the country level, we observe a highly significant positive correlation (p < 0.001) and no meaningful shifts in the country rankings between the measures (See Figure S.5 in the Supplementary material).

Similarly, using the non-top-coded version of the verbal fluency score does not substantially alter the ranking of most countries, with the two measures showing a highly significant positive correlation (p < 0.001). Although the non-top-coded version of the verbal fluency score has little impact on the overall order of countries, it affects the magnitude of the Gini coefficient, particularly for countries with high median scores (see Table S.4 and Figure S.6 in the Supplementary material).

Using the most recent available releases of a wide array of surveys, we mapped inequalities in various health dimensions critical for maintaining an independent life across a uniquely broad geographical scope. While the surveys were conducted in different time periods, which may place some constraints on their comparability, they have been harmonised following the HRS framework, specifically to facilitate cross-country comparisons (Smith, 2021). However, differences in the HGS assessment could introduce biases. Most surveys used the same dynamometer, but a few used different brands and allowed measurements in a seated position (Phillips et al., 2023).

Furthermore, variations in survey designs could limit the comparability of the results. For instance, the Chile-Cog survey only includes older adults aged 60 and older and several countries in SHARE have a low number of observations for the youngest age group (50–54), which may lead to bias in the dispersion of functioning scores. We addressed this issue by conducting country comparisons for the broader 60–75 age group, as the age distributions across countries are similar for this age group. Furthermore, the subsample of 60–75-year-olds reduces sample selection due to the survival of healthier older adults. Another important limitation is that while our sample includes 41 countries, it still only represents a subset of the world’s nations.

Additionally, while the Gini index provides notable advantages as a measure of inequality for the purposes of our study, it also has limitations. The Gini index offers advantages over other indicators, such as the Concentration Index that was commonly used in previous studies focusing on health inequalities (Siegel and Allanson, 2016; Wagstaff et al., 1991), as it measures inequality purely within the health outcomes of interest (i.e., cognitive or physical functioning), making it well-suited for comparisons across countries and survey designs (Steinbeis et al., 2019).

It is, however, crucial to emphasise a major limitation of the Gini index: i.e., its limited ability to represent specific parts of the health distribution. We addressed this issue by including the P80/P20 ratio as a robustness check, specifically focusing on disparities between two specific parts of the distribution, which showed similar results.

Another consequence of this limitation is that as a measure of dispersion, the Gini cannot distinguish whether variation in the distribution occurs on its upper or its lower end. Thus, the Gini index does not fulfil the property of transfer sensitivity, meaning that it is not necessarily more sensitive to health improvements at the lower end of the distribution than to health deterioration at the upper end (Foster et al., 2013).