Main

Climate change and economic inequalities are among the most pressing challenges of our times, and they are interrelated: failure to contain climate change is likely to exacerbate inequalities within and between countries1,2,3,4 and economic inequalities within countries tend to slow the implementation of climate policies5,6. To properly understand the relationship between economic inequality and climate change, sound and timely data about the distribution of greenhouse gases (GHG) emissions between individuals and across the globe are needed. Such information is currently missing. As a matter of fact, researchers, policymakers and civil society struggle to establish even basic facts about which groups of the population contribute to emissions growth, or mitigation. This jeopardizes any efforts towards sustainable lifestyles.

This paper addresses these issues by harnessing recent conceptual and empirical progress in the measurement of income, wealth and GHG emissions. Compared with previous work on global carbon inequality7,8,9,10, this paper presents three major developments in terms of data, methods and scope.

First, the paper uses novel income and wealth inequality data from the World Inequality Database11 to track inequality from the bottom to the top of the distribution. These economic inequality data are combined with GHG footprints from input-output models thanks to a newly assembled set of country-level information on the link between individual emissions, consumption and income in more than 100 countries. The methodology therefore makes it possible to track individual GHG emission levels with more precision than previous longitudinal carbon inequality estimates9. Second, the method developed allows explicitly distinguishing between emissions from private consumption and investments, making it possible to better understand the drivers of emissions among wealthy groups. Third, the paper focuses on the distribution of emissions over the 1990–2019 period, that is, from the first Intergovernmental Panel on Climate Change (IPCC) report to the eve of the Covid-19 pandemic. The three decades saw critical shifts in the distribution of world economic growth12, which have not been systematically studied from the point of view of GHG emissions inequality.

There are two broad approaches to the measurement of global carbon inequality. ‘Bottom-up’ approaches use household-level microdata to produce macroestimates. This is the approach taken by refs. 8,13,14 that mobilize the large set of consumption surveys available from the World Bank Global Consumption Database, as well as additional consumer expenditure surveys done in rich countries. These surveys are linked to Environmental Multi-Regional Input-Output models (EMRIOs) to provide estimates of energy consumption or emissions per consumption group. To the extent that micro-level data are available, this method is the best way to measure global carbon inequality associated with individual ‘consumption’. Given the data-intensive process, this approach has not looked at the evolution of global emissions. Another limitation is that this approach tends to underestimate the consumption levels of the richest groups due to well-documented misreporting and sampling errors15. ‘Top-down’ approaches to the measurement of global carbon inequality use the regularities observed in micro-level data to provide modelled estimates on the basis of elasticity parameters and income or consumption inequality distributions. This is the approach taken by refs. 7,9,10,16. These studies typically use one single elasticity for all countries, which limits the precision of country-level estimates. Another limitation of both top-down and bottom-up approaches is that they do not treat investment-related emissions particularly well.

The present paper builds on the strengths of top-down and bottom-up approaches and offers novel developments. By mobilizing country-level elasticities from over a hundred countries, the paper departs from previous top-down approaches. By focusing on the 1990–2019 period, the paper adds historical depth to single-year bottom-up studies, and by distinguishing between emissions from personal consumption and from investments, it sheds new light on the dynamics of emissions, in particular among top groups.

The general approach followed here can be summarized as follows: using EMRIOs, I obtain country-level GHG emissions for the household sector, the investment sector and the government sector across countries (emissions are net of imports and exports embedded in goods and services traded with the rest of the world). These emissions are distributed to individuals in each country using country-level data on the elasticity of emissions and consumption, income and wealth. A variety of alternative estimation strategies are tested and it appears that the key results are robust to a large range of parametric assumptions on the relationship between emissions, income, consumption and wealth. To be sure, a lot remains to be learned and debated about the link between individual emissions and wealth. As in any excercise of this sort, the statistical reconstructions presented below should be analyzed with caution.

Results

Carbon emission inequalities within regions

Global average per-capita emissions reached about 6 tonnes of carbon-dioxyde equivalent (tCO2e) in 2019. To have high chances of staying below +1.5 °C global temperature increase, average per-capita emissions should be 1.9 tCO2e between now and 2050 (that is, the equivalent of an economy-class round-trip flight between London and New York) and zero afterwards (Supplementary Information section 3).

Inequalities in average per-capita emissions ‘between’ world regions remain large, as shown in Extended Data Fig. 1. On top of these gaps, important inequalities in carbon footprints are observed within regions. Figure 1 presents the carbon footprints of the bottom 50% of emitters, the middle 40% and the top 10% of the population within regions according to my benchmark results. Emission levels and shares for other groups are presented in Supplementary Information (section 7).

In East Asia, the poorest 50% emit on average 2.9 tCO2e per annum, while the middle 40% emit nearly 8 t, and the top 10% almost 40 t. This contrasts sharply with North America, where the bottom 50% emit fewer than 10 t, the middle 40% around 22 t and the top 10% around 69 tCO2e. This in turn can be contrasted with the emissions in Europe, where the bottom 50% emit 5 t, the middle 40% around 10.5 tCO2e and the top 10% around 30 tCO2e. Emission levels in South and Southeast Asia are notably lower than in the other regions, from around 1 tCO2e for the bottom 50% to 11 t on average for the top 10%.

It is striking that the poorest half of the population in the United States has emission levels comparable with the European middle 40%, despite being almost twice as poor as this group in purchasing power parity terms17. Conversely, the top 10% of the population in East Asia emits notably more than its European counterpart (40 tCO2e vs 29 tCO2e, respectively). It also appears that Russia and Central Asia have an emissions distribution broadly similar to that of Europe, but with higher top 10% emissions (mainly due to higher income and wealth inequalities in Russia and Central Asia) and lower bottom 50% emissions. Sub-Saharan Africa lags behind, with the bottom 50% emitting around 0.5 t per capita and per year, and the top 10% emitting around 7.5 t.

Global carbon inequality between individuals

Figure 2 presents the inequality of carbon emissions between individuals at the world level. The global bottom 50% emit on average 1.4 tCO2e per year and contribute to 11.5% of the total. The middle 40% emit 6.1 t on average, making up 40.5% of the total. The top 10% emit 28.7 t (48% of the total). The top 1% emits 101 t (16.9% of the total). Global carbon emissions inequality thus appears to be great: close to half of all emissions are released by one-tenth of the global population, and just one-hundredth of the world population (77 million individuals) emits about 50% more than the entire bottom half of the population (3.8 billion individuals).

The evolution of individual carbon emission inequalities

How has global emissions inequality changed over the past few decades? In Fig. 3a, global polluters are ranked from the least emitting to the highest on the X axis, and their per-capita emissions growth rate between 1990 and 2019 is presented on the Y axis. Since 1990, average global emissions per capita grew by 2.3% (and overall emissions grew by about 50%, see Supplementary Information Table 6.1). The per-capita emissions of the bottom 50% grew faster than the average (26%), while those of the middle 40% as a whole was negative (−1.2%), and some percentiles of the global distribution actually saw a reduction in their emissions of between 5 and 25%. Per-capita emissions of the top 1% emissions grew by 26% and top 0.01% emissions by 80%. One striking result shown in Fig. 3a is the reduction in the emissions of about 5–15% for percentiles p75 to p95. This segment of the world population largely corresponds to the lower- and middle-income groups of the rich countries and contrasts with the emissions of the top 1%, which have markedly increased.

Per-capita emissions matter but understanding the contribution of each group to the overall share of total emissions growth is also crucial. The bottom half of the global population actually contributed only 16% of the growth in emissions observed since then, while the top 1% (77 million individuals in 2019) was responsible for 23% of total emissions growth. The top 0.1% (7.7 million individuals in 2019) contributed about two-thirds of the entire growth in emissions associated with the poorest half of the global population (3,855 million individuals in 2019). Supplementary Information Table 7.1 presents the evolution of the Theil and Gini indices of global emissions inequality.

Global carbon inequality dynamics are governed by two forces: the evolution of average emission levels ‘between’ countries and the evolution of emission inequalities ‘within’ countries. Which of these two forces has been driving the dynamics of global carbon inequality over the past few decades? Fig. 3b compares the share of global emissions due to within-country differences with the between-country differences using a Theil-index decomposition. In 1990, most global carbon inequality (62%) was due to differences between countries in my benchmark estimates: back then, the average citizen of a rich country polluted unequivocally more than the rest of the world, and income inequalities within countries were on average lower across the globe than today. The situation has entirely reversed in 30 years. Within-country emission inequalities now account for nearly two-thirds of global emissions inequality. To be clear: this does not mean that important inequalities in emissions between countries and regions have disappeared. On the contrary, it means that on top of the great inter-national inequality in carbon emissions, there are also even greater emission inequalities between individuals within countries. To provide another insight into this result, Fig. 4ab presents a geographical breakdown of global emitter groups: it reveals that top global emitters come from all world regions.

Investments and the carbon footprints of wealthy individuals

In the framework developed, personal carbon footprints can be split into emissions generated by private consumption, investments and government spending. Consumption-related emissions come from the carbon released by the direct use of energy (for example, fuel in a car) or its indirect use (for example, energy embedded in the production of goods and services consumed by individuals). Investment-related emissions are emissions associated with choices made by capital owners about investments in the production process (that is, emissions involved in the construction of machines, factories and so on). Emissions from government spending correspond to collective consumption expenditure or investments (government administration, public roads, defense, etc.).

Focusing on the breakdown between consumption and investment emissions, I find that the bulk of the emissions generated by the global top 1% comes from their investments rather than their consumption (Fig. 5b) (over 70% in 2019 in the benchmark scenario). The weight of investments in the per-capita footprint of the top groups has been rising since the 1990s. This appears to be due to the rise in overall emissions associated with investments over the period (Supplementary Table 1.1) as well as to the rise in wealth inequality: wealth and investments are more concentrated today than they were in 1990 in many countries.

Discussion

The results presented above reveal the very highly skewed concentration of individual carbon emissions that characterizes the contemporary global economy: while one-tenth of the global population is responsible for nearly half of all emissions, half of the population emits less than 12% of it. Seen in perspective, carbon inequalities are lower than income and wealth inequalities (the global top 10% of earners captures 52% of total income and the global top 10% of wealth owners owns three-quarters of total wealth18). Global carbon inequalities nonetheless remain quite large today and show no sign of clear decline despite some convergence observed between countries.

Focusing on 1990-2019 dynamics, the increased emissions by top global emitters since 1990 are particularly striking when compared with the emission trajectories of other population groups. Indeed, the per-capita emissions of the poorest 50% in Europe and the United States appear to have dropped by around –25-30% since 1990 (see Appendix Table 7.2). These reductions are the result of the combined effect of compressed wages and consumption and a reduced national per-capita footprint in most rich countries, driven by climate and energy policies and efficiency gains in industrial processes. Consequently, a large part of the population in rich countries already appears to be near 2030 national climate targets when these are expressed in per-capita terms. Nationally Determined Contributions (NDCs) established under the rubric of the Paris Agreement imply a per-capita target of around 10 t of CO2e in the United States in 2030 and around 5t for European countries in my benchmark results. In the United States and in some European countries, I find that the bottom 50% of the population is relatively close or even meets these 2030 targets (Fig. 6 and Supplementary Information section 8.1). This is not the case for the middle 40% and top 10% of the income distribution in these countries. In the United States, the top 10% would have to reduce its average per-capita emissions by 86% to reach the 2030 target, with the value being 81% in France according to the benchmark estimates.

In emerging and developing countries, 2030 climate targets imply an increase in average per-capita emissions rather than a reduction. In these countries, however, inequality also matters a lot: in China and India, emissions of the bottom 90% of the population are below the target, while those of the wealthiest 10% are already well above it. In China, the richest 10% of the population would have to reduce its emissions by more than 70% to reach the 2030 target, and the reduction needed is over 50% for India in the benchmark estimates (Fig. 6 and Supplementary Information section 8).

To be clear, no country currently envisages the enforcement of strict per-capita targets to meet its 2030 objectives. Nonetheless, the gaps between individual emission levels and the implied national target raise important questions about the design of climate and sustainability policies in the years to come: how do we ensure that regulations, tax instruments and other climate policies effectively address the emissions of the high emitters?

There is no straightforward answer to such questions, but it appears that climate policies over the past decades have often targeted low-income and low-emitter groups disproportionately, while leaving high emitters relatively unaffected. The trends documented in this paper tend to support this view. In fact, key climate policy instruments (such as carbon taxes, for instance) have done little to address the vast inequalities in carbon footprints, and may have exacerbated them in some countries. Carbon taxes have been found to place a disproportionate burden on low-income and low-emitter groups19,20,21, while the carbon price signal for high and wealthy emitters may be too low to force changes in consumption (or investment) patterns among wealthy individuals.

Extended Data Fig. 2 presents several options to better integrate inequality in the design of climate policies. Focusing on the specific issue of the carbon content of investments, it appears that progressive carbon tax systems could be helpful to accelerate decarbonization. To design progressive carbon tax systems, one option is to combine carbon pricing with cash transfers for certain categories of the population, as has been done in British Columbia (Canada)5. Another option is to make carbon tax rates increase with emission levels. This could potentially be achieved via a combination of tax instruments, focusing on consumers as well as on investors in carbon intensive activities. Today, states typically do not impose taxes or regulations on the basis of the pollution content of asset portfolios or of investments. This can be seen as paradoxical given that investors have a variety of options for investing their wealth, and it stands in stark contrast with low-income consumers who do not always have alternatives, in the short run, to using fossil fuels, but who must pay carbon taxes.

Using the data constructed for this paper, it appears that the global top 1% would contribute to about 40% of total revenues from an additional carbon tax focusing on the carbon content of investments and the top 10% three-quarters of the total. With a tax rate r = 0 for annual investments with a carbon content below 5 tCO2e per capita and r > 0 for investments with a carbon content above this threshold, close to 100% of the tax would fall on the top 10% of the global population. Under this schedule, the bottom 77% of the US population, the bottom 90% of the European population and the bottom 99.5% of the Sub-Saharan African population would not pay the tax at all (Supplementary Information Table 8.3). Such a tax could be used as a top-up mechanism to make overall carbontax systems more progressive and to raise additional revenues to invest in low-carbon infrastructures, or to compensate losers in the transition. The technical and economic conditions under which policies targeting the carbon content of investments are developed is a matter for further research. In any case, more transparency on individual carbon footprints and on which socio-economic groups contribute to decarbonization efforts will be paramount to guarantee a just transition.

Methods

Environmental input-output data

The most straightforward way to obtain internationally comparable direct and indirect emission levels of individuals is the input-output (I-O) framework applied to the environment framework developed in ref. 22. The benchmark I-O data source used in this paper is the Global Carbon Project23. The paper also relied on the EORA dataset24. Emissions include all GHGs except those related to land-use change. For details on the construction of I-O carbon aggregate series used in this study, see Supplementary Information section 1 and ref. 25.

Income and wealth inequality data

The past two decades were marked by breakthroughs in our capacities to monitor income and wealth inequality within countries18,26, which the paper builds upon. The standard source of information for tracking inequality within countries is household surveys, which typically fail to properly measure incomes and wealth at the top of the distribution, and are usually not consistent with macroeconomic totals12,27, making cross-country comparisons difficult. The Distributional National Accounts (DINA) methodology11,28 addresses these issues by systematically combining household surveys with additional sources of information (including, in particular, administrative tax data and national accounts).

This study relied on the DINA project, which provided detailed income and wealth inequality series for 174 countries for the 1990–2019 period, that is, for around 97% of the world population and over 97% of global gross domestic product. The general guidelines and methods underlying these data series are described in the DINA Guidelines11 (see also Supplementary Information section 4).

Elasticity between carbon emissions and consumption or income

Data on individual emission inequalities have been produced for several countries and years by researchers using input-output analysis applied to the environment and household surveys. Available literature typically finds that carbon emissions associated with individual consumption depend on several factors, including income, household location, energy conversion technologies, occupation status, habits, age, national regulations and energy mixes14,29,30,31,32,33,34,35,36,37. While non-income factors play a role in determining individual emission levels, income retains a very large role in explaining variance in emissions between households.

Studies measuring the ‘elasticity’ of individual carbon emissions (or the strength of the relationship between rising individual income and CO2 emissions) are presented in Supplementary Information section 5. These studies find that the elasticity of household consumption to emissions (in a model of the form e = k × cα, where e is the level of emissions, c is consumption, α is the elasticity and k is a constant) typically falls in the 0.9–1.1 range, while the elasticity of household income to emissions typically falls in the 0.5–0.7 range (Supplementary Information Table 5.1). This paper mobilizes these country-level elasticities, now available for most countries, to produce relatively more fine-grained modelled estimates of the distribution of emissions than earlier top-down studies.

Distributing emissions among individuals

National-level distributions of income, wealth or carbon emitters were broken down into 99 percentile groups and 28 smaller fractiles within the top percentile. Average per-capita emissions at percentile p in a given year and country are defined as

$${E}_{p}^{\mathrm{tot}}={E}_{p}^{\mathrm{cons}}+{E}_{p}^{\mathrm{inv}}+{E}_{p}^{\mathrm{gov}}$$
(1)

where $${E}_{p}^{\mathrm{cons}}$$, $${E}_{p}^{\mathrm{inv}}$$ and $${E}_{p}^{\mathrm{gov}}$$ are individual average footprints at percentile p, associated with household consumption, private investment and public spending, respectively. More precisely:

$${E}_{p}^{\mathrm{cons}}=f({E}^{\mathrm{cons}},{y}_{p},\alpha )$$
(2)
$${E}_{p}^{\mathrm{inv}}=f({E}^{\mathrm{inv}},{w}_{p},\gamma )$$
(3)
$${E}_{p}^{\mathrm{gov}}=f({E}^{\mathrm{gov}},{y}_{p},\delta )$$
(4)

where Econs is the average carbon footprint associated with a unit of consumption in the country, yp is the average income level of individuals in percentile p, α is the elasticity of household consumption carbon emissions to income (in a model of the form $${E}_{p}^{\mathrm{cons}}=k{E}^{\mathrm{cons}}\times {y}_{p}^{\alpha }$$); Einv is the average emissions level associated with fixed capital formation, wp is the average wealth level of individuals in percentile p, γ is the elasticity of wealth to investment emissions; Egov is the average emission level of the government sector (associated with in-kind redistribution) and δ is the elasticity of government emissions to individual income.

The benchmark results presented above are based on α values available from country-level studies based on microdata. I also tested a variety of α values for each country from Supplementary Information Table 5.1.

Fitting the model with observed γ is a challenging task given how few studies of the topic exist. The elasticity of asset ownership is reported to be near unity38. Limited available evidence suggests that the distribution of emissions associated with wealth ownership is close to proportional to the distribution of wealth ownership (see also Supplementary Information section 5). The elasticity of asset ownership is reported to be near unity in ref. 38 and this finding tends to be corroborated by the data produced by ref. 39.

The benchmark scenario is based on δ = 0. This amounts to distributing government collective consumption expenditure equally to individuals as a lump-sum. This choice tends to minimize inequality in carbon emissions between income groups at the country level. In alternative scenarios, I distributed emissions in proportion to individuals’ private consumption.

Besides the benchmark scenario, I produced results for the following set of parameters: α = (0.4; 0.5; 0.6; 0.7; 0.8); γ = (0.9; 1; 1.1); δ = (0; 1). Extreme scenario bounds presented in the Figures are based on extreme bounds observed in available country-level data, that is, α = (0.4; 0.8) or γ = (0.9; 1.1). In all countries, it is assumed that emissions are split equally within households.

Robustness checks

Supplementary Information section 7 provides results for different parametric assumptions at the global, regional and country levels. The main findings of this paper appear to be robust to a wide set of different assumptions. In an extreme lower-bound scenario (in which all countries would have the lowest empirically observed α value), I find that the global top 10% share of emissions nears 45% in 2019 (vs 48% in the benchmark scenario). In an extreme upper-bound scenario (in which all countries would have the highest empirically observed α value), I find that the global top 10% share is 51%. Setting different γ parameters affects results at the top of the distribution, although in a moderate way: with γ = 0.9 (and using empirically observed α values), the global top 10% share is equal to 46% in 2019. With γ = 1.1, the global top 10% share is equal to 50% in 2019. Opting for δ = 1 yields a global top 10% share of around 50% and a bottom 50% share near 10%. It also appears that setting δ = 1 has a fairly limited impact on bottom and top groups’ overall emissions, as can be seen in Supplementary Information Table 7.2, given that overall government emissions remain relatively low as compared with private consumption and investments.

Global dynamics between 1990 and 2019 also appear to be robust across these different scenarios and are not particularly sensitive to changes in parameter values within plausible bounds, as presented in Figs. 1, 2 or 3. Changes in α values over time also seem to have little impact on global results, as illustrated in Fig. 3b: if α had decreased in all countries from 0.8 to 0.4 between 1990 and 2019 (that is, if the wealthy had done much more decarbonization efforts than the rest of the population, per dollar spent), global emissions inequality would still be essentially driven by within-country dynamics today. Let me stress at the outset that, given the nature of the reconstruction exercise presented above, within-country percentile-level estimates should be interpreted with care: a lot remains to be done by governments to improve the quality of distributional and environmental statistics. This novel set of estimates is as much a progress in our understanding of global carbon inequalities as a mapping of the many data and conceptual gaps which will have to be addressed in further research.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.