The COVID-19 pandemic began in April 2019. 17% of UK workers lost their jobs by April 2020. Women were 4.8 percentage point more likely to lose their jobs than men. The average income of a South African household with white members was 5.6 times higher than that of a household made up of Africans in 2017. (Adjusted for the size and composition households, ‘African’ is an officially recognized racial category in South Africa). A 2018 study in the United States found that those born to parents in the lower half of the distribution of ‘years of education’ had only 13% chance of reaching the top quartile in their generation’s schooling.
The differences between these groups are not determined by how hard people study or work, how much they save, or how responsible they are. The gaps between the groups are purely due to certain characteristics that individuals cannot control, such as gender, race, and upbringing. These are examples of what economists refer to as inequality of opportunity.
Multiple lines of evidence support the idea that inequality of opportunity is the most damaging component of inequality. However, economists and other researchers fail to measure its impact on the measurement of inequality of income, education, employment, or any other outcome.
This requires as much information as possible about people’s lives, including their race, ethnicity, where they live, and the education of their parents. Even if such data are available, there is not widespread agreement about how to measure the inequality that exists in a population due to inequality of opportunities. This must change.
Inequality of opportunity is a key topic in economic discourse. Research should focus on collecting data that can be used to make credible estimates. Economists and other professionals should develop better methods and tools to create those estimates.
Since the Enlightenment of the seventeenth and eighteenth century, the ideal of equal opportunity has been a central part Western political discourse. It is still prominent today.
For example, US President Franklin Roosevelt stated in his speech at Little Rock, Arkansas on 10 June 1936 that “equality of individual ability has never existed” and that “equality of opportunity must still be sought.” In his 1971 treatise A Theory of Justice (which was highly influential), John Rawls argued that fair equality of opportunities, which is the idea that all citizens should have equal access to goods, services, and employment opportunities, is one of the two principles of social justice. The other principle is that everyone should enjoy the same freedoms as everyone else.
The concept was still intangible until economists like Marc Fleurbaey and John Roemer attempted to formalize it in the 1990s4-6. These economists suggested that each of the factors that determine a social outcome, such as earnings, could be broken down into factors that an individual can or cannot control. The first was defined as an individual’s efforts and the second as the circumstances. This simple formula provided the basis for an economic theory on inequality of opportunity.
The results of subsequent research, mainly in behavioral economics, have shown that not only philosophers and economists care about the distinction between circumstances or efforts. Since the early 2000s, people have been given real money in multiple field and laboratory experiments. They are instructed to share it with others, as well as among themselves.
Only a small number of people will make decisions in such circumstances that are consistent with the expectations of ‘Homoeconomicus’. This is a figurative species, which is characterised, amongst other things, as being purely self-interested. For example, if a participant receives US$100 they are more likely than to give away $1 or keep $99. Furthermore, participants who propose an unfair allocation of money are more likely to give away $1 or $50, rather than giving away $1 and keeping $99.
In other words, people (and some animals) have an intrinsic preference to fairness and equity.
Economists looked deeper into these behaviors and discovered that people are most offended by inequality that results from external factors.
For example, in a 2010 study, 238 students were hired by economists to type as many words possible within a 10- or 30-minute time frame8. The task could be either a short or a longer one. The number of words they correctly identified was the basis for their pay. However, random numbers were used to determine which students were paid $0.08 and $0.16 respectively.
After students were divided into pairs and given information about the work time, word output, and payment, each student was able to propose a redistribution of earnings. Participants who were randomly assigned a low wage were most likely to be compensated. A smaller number chose to compensate for work’s duration and quality, which were both considered within the individual’s reach8.
It is becoming increasingly clear that inequality of opportunity can have a negative impact on other social outcomes such as economic growth.
Researchers have studied the relationship between income inequality, factors like violent crime9 and disease and death10 since the 1990s. These studies have generally shown that inequality is associated with poorer health outcomes and worse social outcomes. However, the evidence linking inequality to growth is less clear. Different data sets and methods have led to inconsistent results11,12.
Researchers have separated the inequality of opportunity from growth over the past decade to make the associations between inequality and growth clearer.
For example, a study found no statistically significant relationship between economic growth and inequality in 26 US states from 1970 to 2000, when total income inequality was taken into account. However, total inequality could be broken down into two components: one that was due predetermined circumstances (inequality of opportunity), and the other that was dueto other factors. The former had a significant negative effect on growth13.
Another study supports the notion that when large groups are denied productive opportunities – simply because of their personal circumstances – the result is lower economic efficiency and dynamism. For example, a 2019 study examined the percentage of white men among the US’ doctors and lawyers. This figure fell from 94% in 1960, to 62% in 2010. This study found that 20-40% of gross domestic product per capita growth over the 50-year period could be attributed to this decline (and similar declines within other professions). These gains may have been due to a better allocation of talent. Women and Black men (the other two groups) took up more opportunities that were previously closed to them15.
If inequality of opportunity is the ‘active ingredient” of inequality, which is the most disliked and causes the most severe social effects, why are there still so few measures?
“Opportunity” refers to the number of opportunities that are available to a person. It is therefore difficult to quantify. However, economists have developed relatively easy ways to measure inequality of opportunities based on the work of Fleurbaey and Van de gaer.
These measures attempt to quantify the extent of inequality in income or education due to factors beyond the control of people. They also account for the fact that individuals’ efforts can be influenced by their environment16-18.
To obtain such information, you need to have as much detail as possible about each individual in the sample. It is relatively easy to obtain data on race, ethnicity, gender, and place of origin. However, to quantify inequality of opportunity, it is important to have detailed and long-term information about a person’s entire family history, including their education, income, occupations, and parenting behaviors.
These data are not common in low- or middle-income countries. However, they do exist in some wealthy ones. However, they are available for high-income countries such as the United States of America, Germany, the United Kingdom, and most of Scandinavia.
Two types of data that are considered the best should be collected in order to allow researchers and policymakers to better understand the uneven opportunities faced by people around the globe.
The first is based on longitudinal, detailed household surveys. The US Panel Study of Income Dynamics was launched in 1968. Germany’s Socioeconomic Panel has been running since 1984. These panels offer rich information about the parenting habits and parents of adults today and enable economists to connect information across generations. Although similar initiatives have been started in emerging economies like Mexico and Indonesia, they are very rare.
The second type comes from administrative data sets, which connect personal identifiers across generations and across various aspects of people’s life: educational outcomes can be linked with employment and health histories as well as social-security contributions and tax payments.
Researchers have to be careful about confidentiality and privacy when they have access to sensitive and rich data. These data are accessible to researchers in Norway and Denmark, but they have to be subject to certain procedural restrictions that protect privacy. Similar developments are also being made in Chile and elsewhere.
It is easy to see how global data collection on individuals could be improved with enough investment.
Economists cannot accurately assess inequality of opportunity even with the most accurate data. They can only do this if they know exactly how variables should be used in order to divide a population into groups that are similar.
Theoretically, all circumstance variables should be used when data is available for a whole population. Some variables may not be observed even though they are available for the entire population. Researchers may have information about the parents of those who went to university but not the ranking of universities. Some circumstances cannot be ignored, so any estimation of inequality of opportunity is likely to be underestimated.
However, if data are only available for a small sample of the population (as it is often the case), over-fitting can occur when too many variables are used to divide the sample. Subgroups with fewer members tend to have noisy estimates of their group-specific characteristics. This sample error can cause an increase in the percentage of variation due to inequality of opportunity. These two contradictory biases make it difficult for researchers to decide how to best partition a population by circumstance variables.
Machine-learning techniques19-20 are a promising option. An algorithm will attempt to split a set of data into two groups based on the statistical differences between them. The process is repeated for each subgroup until there are no more significant differences (for some standard level or statistical significance).
My colleagues and I conducted an unpublished analysis of a sample consisting of just over 6 000 South African households. The income data was collected by the National Income Dynamics Study. The sample consisted of people who identified themselves as being one of the following four groups: Africans, Coloured (a South African recognized racial category), Asian/Indian or white.
An algorithm was used to divide the sample. Each group should contain at least 100 observations. 1% was used as the threshold for statistical significance. These parameters were used to divide the sample into ten types or subgroups.
We then gave each person in each subgroup an average income and computed inequality using this smoothed distribution. This ensures that there is no inequality of opportunity between the subgroups.
This Gini coefficient of 0.61 estimates that there is an inequality of opportunity between 66% to 74% of South Africa’s overall income inequality. The Gini coefficient is a measure of inequality in household income distribution. Lower values indicate a more equitable distribution. Using our lower estimate, 66% of 0.61 is 0.40, which is similar to the latest Gini coefficient reported by the World Bank for the United States (0.42 in 2019; see go.nature.com/3xpjmwp). This means that the inequality among these ten subgroups is almost equal to the overall inequality in the United States. The share of the total ranges between 39% and 48% using a more conservative measure (the median logarithmic deviation). A 2015 South African study using 54 types of data found that inequality of opportunity was responsible for just 17-24% of the average logarithmic deviation21.
To determine if this machine-learning approach truly represents the level of inequality of opportunity in a society, it must be tested in multiple settings. Some may be concerned that some algorithms could be considered methodological “black boxes”. A machine-learning approach is difficult to use in a situation where no causal hypothesis has been tested and researchers seek the most efficient way to use the data available to measure inequality of opportunities.
Inequality of opportunity is the channel through which inequality is transmitted from generation to generation. New computational statistical methods could be used to quantify inequality in countries and regions around the globe, if combined with better data collection efforts.
This information could be used to complement measures of economic mobility, which measure the relationship between an adult’s earnings and those of their parents. These measures rely only on one circumstance variable: parental income. It could also be used to supplement measures of overall inequality, such as income, wealth, or education.
Opportunities and outcomes can be best viewed as two sides of one coin. A family’s current outcomes shape the future opportunities of their children, and these opportunities in turn help determine the children’s future outcomes.
