Socioeconomic inequality in longevity is larger than we thought

It is well established that individuals with higher socioeconomic status live longer than those with lower socioeconomic status, but most evidence on this is based on a single indicator of status. This column uses Danish population data and machine learning methods to show that combining multiple socioeconomic indicators (income, education, wealth, occupation, and IQ) reveals a much steeper socioeconomic gradient in life expectancy than standard single-indicator measures suggest.

It is well established that individuals with higher socioeconomic status (SES) live longer than those with lower SES, and that these gaps have widened in many countries (Chetty et al. 2016, Kinge et al. 2019, Milligan and Schirle 2021). This inequality in longevity has implications for the sustainability and equity of social security, retirement ages, and healthcare systems (National Academies 2015, Auerbach et al. 2017) and features prominently in policy discussions (OECD 2018, 2022, United Nations 2022, 2023).

Most evidence on longevity inequality measures SES using a single indicator – typically, education or income. But SES is inherently multidimensional – for example, a person can have high education but low wealth, or high income but a low prestige occupation. If the relevant ‘social position’ for health and longevity is a combination of several dimensions, single-indicator gradients will tend to be lower bounds on the true SES gradient in life expectancy.

In a recent paper (Bingley et al. 2025), we use population-wide Danish administrative data and machine-learning methods to construct a comprehensive measure of SES-related inequality in life expectancy. We follow all individuals in Danish birth cohorts 1942–44 from age 40 to 78 and measure SES at age 38 using income, education, wealth, occupation, and (for men) IQ from military conscription.

The extent of inequality in longevity

Figure 1 plots cohort life expectancy at age 40 against an SES percentile rank. For the multi-factor curve, we use a machine-learning model to combine all available SES indicators into a predicted longevity rank (1–100). For comparison, we repeat the exact same procedure using each indicator on its own, so multi-factor and single-factor gradients are compared on equal terms.

For men, predicted life expectancy at age 40 rises from about 65 years at the bottom of the multi-factor distribution to 89 years at the top – a gap of 24 years. This gap is substantially larger than gaps implied by common single-indicator measures. Using education alone yields a gap of about 9 years, while income alone yields about 17 years. The other single indicators are similar in magnitude: IQ about 9 years, wealth about 17 years, and occupation about 19 years.

The same message holds if we focus on the slope rather than the extremes. Moving up ten percentiles in the income-based rank is associated with an increase in life expectancy of 1.2 years, while the same move in the multi-factor rank corresponds to 1.7 years. In other words, the gradient is about 50% steeper when we combine indicators. Relative to education alone, the slope is more than twice as steep (a 120% increase).

For women, the difference between multi-factor and single-factor estimates is even more pronounced (despite having one fewer SES indicator available). Women live longer on average than men, but the gap between low-SES and high-SES women is still about 23 years using the multi-factor measure – almost the same as for men. By contrast, the gap is only about 9 years using education, income, or occupation alone, and about 12 years using wealth alone.

Figure 1 Life expectancy at age 40 by SES

(a) Men

(b) Women

Note: Cohort life expectancy at age 40 by single-factor and multi-factor measures of socioeconomic status. Source: Bingley et al. (2025)

Taken together, combining indicators increases the steepness of the SES gradient by 50–150% compared with income or education alone.

Why combining SES indicators changes the picture

The reason the multi-factor gradient is so much steeper is that the individuals at the very top and bottom of the longevity distribution are typically those who are consistently advantaged or disadvantaged across several SES dimensions at once. Being ‘top’ in one indicator is often not enough: high income with low wealth, or high education with a low-prestige occupation, tends to place individuals closer to the middle of the predicted longevity distribution than to its extremes.

This pattern implies that single indicators – especially education or income – systematically miss an important part of socioeconomic stratification. They capture some of the longevity gradient, but they do a poorer job of identifying individuals who accumulate advantages (or disadvantages) across domains. The result is that single indicators produce a flatter gradient and a smaller implied gap between ‘high SES’ and ‘low SES’ groups.

Which indicators matter most?

When we assess the marginal contribution of each indicator, wealth stands out as the single most informative factor; excluding wealth from the multi-factor model causes the largest drop in the measured inequality.

At the same time, the analysis also highlights a practical takeaway: even when only a subset of indicators is available, combining what you have can deliver large gains. In particular, combining the two most commonly available measures – income and education – improves the measurement of longevity inequality dramatically relative to using either one in isolation.

Implications for policy and measurement

Two implications follow. First, life expectancy gradients derived from single indicators should be interpreted as lower bounds on socioeconomic inequality in longevity. This distinction is critical for accurately monitoring, reporting, and comparing inequality across countries and time.

Second, distributional assessments of pension and retirement-age reforms should, where possible, use multidimensional SES measures. If longevity differences are much larger than suggested by education or income alone, reforms that look neutral on average may be substantially regressive in terms of lifetime benefit receipt.

Source : VOXeu